JRDB Leaderboard

2D Detection Submissions

Name

OSPA_IoU↓

AP_0.3 ↑

AP_0.5 ↑

AP_0.7 ↑

MMPAT_CVPR21

0.592

75.916%

67.883%

44.897%

Y. He, W. Yu, J. Han, X. Wei, X. Hong and Y. Gong Know Your Surroundings: Panoramic Multi-Object Tracking by Multimodality Collaboration in CVPRW, 2021

Details

Name	MMPAT_CVPR21
Submission Date	2021-12-07 15:58:55+00:00
Abstract	Panoramic image detection and tracking using both image and point clouds. For detection, we adopt the Cascade rcnn as our object detector. The backbone of the detector is ResNet 101. The detector is pre-trained on the COCO dataset and then fine tuned on the JRDB dataset. During training, we employ data augment methods such as mixup, random crop, multiscale training to augment the dataset. During inference, we employ horizontal flip and softnms techniques to generate more robust resutls.
Publication title	Know Your Surroundings: Panoramic Multi-Object Tracking by Multimodality Collaboration	Publication authors:	Y. He, W. Yu, J. Han, X. Wei, X. Hong and Y. Gong
Publication venue and year	CVPRW, 2021	Publication URL:	https://arxiv.org/abs/2105.14683
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.592368	0.334181	0.258188
cubberly-auditorium-2019-04-22_1	0.516372	0.260592	0.255779
discovery-walk-2019-02-28_0	0.546667	0.271201	0.275466
discovery-walk-2019-02-28_1	0.532616	0.267633	0.264983
food-trucks-2019-02-12_0	0.514901	0.314436	0.200464
gates-ai-lab-2019-04-17_0	0.510104	0.137081	0.373023
gates-basement-elevators-2019-01-17_0	0.494435	0.299393	0.195042
gates-foyer-2019-01-17_0	0.446649	0.177997	0.268652
gates-to-clark-2019-02-28_0	0.494765	0.205973	0.288791
hewlett-class-2019-01-23_0	0.432676	0.13251	0.300166
hewlett-class-2019-01-23_1	0.357891	0.156485	0.201407
huang-2-2019-01-25_1	0.57113	0.278081	0.293049
huang-intersection-2019-01-22_0	0.661279	0.343251	0.318028
indoor-coupa-cafe-2019-02-06_0	0.697011	0.506601	0.190411
lomita-serra-intersection-2019-01-30_0	0.640248	0.342354	0.297895
meyer-green-2019-03-16_1	0.703313	0.439837	0.263476
nvidia-aud-2019-01-25_0	0.61862	0.370124	0.248496
nvidia-aud-2019-04-18_1	0.466581	0.20193	0.264651
nvidia-aud-2019-04-18_2	0.506626	0.148189	0.358437
outdoor-coupa-cafe-2019-02-06_0	0.759582	0.547651	0.211931
quarry-road-2019-02-28_0	0.631906	0.149522	0.482384
serra-street-2019-01-30_0	0.719977	0.468965	0.251012
stlc-111-2019-04-19_1	0.352728	0.098974	0.253754
stlc-111-2019-04-19_2	0.378001	0.11886	0.259141
tressider-2019-03-16_2	0.656735	0.362998	0.293737
tressider-2019-04-26_0	0.71067	0.48345	0.22722
tressider-2019-04-26_1	0.675365	0.512639	0.162726
tressider-2019-04-26_3	0.651699	0.379319	0.27238

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.759155	0.678829	0.448973
cubberly-auditorium-2019-04-22_1	0.783587	0.733309	0.486899
discovery-walk-2019-02-28_0	0.896124	0.825952	0.550954
discovery-walk-2019-02-28_1	0.854891	0.803498	0.607199
food-trucks-2019-02-12_0	0.842216	0.789975	0.576813
gates-ai-lab-2019-04-17_0	0.80933	0.758997	0.61905
gates-basement-elevators-2019-01-17_0	0.805602	0.798955	0.628118
gates-foyer-2019-01-17_0	0.909822	0.835772	0.695075
gates-to-clark-2019-02-28_0	0.810159	0.760957	0.647977
hewlett-class-2019-01-23_0	0.877526	0.861023	0.678432
hewlett-class-2019-01-23_1	0.950697	0.907767	0.746809
huang-2-2019-01-25_1	0.707445	0.632578	0.3902
huang-intersection-2019-01-22_0	0.78833	0.742469	0.480125
indoor-coupa-cafe-2019-02-06_0	0.735423	0.633404	0.405672
lomita-serra-intersection-2019-01-30_0	0.812648	0.719845	0.518871
meyer-green-2019-03-16_1	0.617408	0.502327	0.278515
nvidia-aud-2019-01-25_0	0.721449	0.6284	0.361653
nvidia-aud-2019-04-18_1	0.847472	0.784675	0.637034
nvidia-aud-2019-04-18_2	0.841472	0.781549	0.521843
outdoor-coupa-cafe-2019-02-06_0	0.653672	0.53751	0.323386
quarry-road-2019-02-28_0	0.766861	0.713112	0.494631
serra-street-2019-01-30_0	0.778041	0.673061	0.37536
stlc-111-2019-04-19_1	0.868494	0.834388	0.674827
stlc-111-2019-04-19_2	0.879542	0.835003	0.712373
tressider-2019-03-16_2	0.662335	0.586687	0.384583
tressider-2019-04-26_0	0.729313	0.611034	0.321686
tressider-2019-04-26_1	0.756105	0.674612	0.471268
tressider-2019-04-26_3	0.744963	0.644879	0.361184

T_HJ

0.607

75.766%

68.097%

45.121%

Anonymous Submission

Details

Name	T_HJ
Submission Date	2021-12-16 22:26:45+00:00
Abstract	Using the stitched image, we trained the type of R-CNN. The backbone of the detector is ResNet 101. The detector is pre-trained on the COCO dataset and then fine tuned on the JRDB dataset. During training, we employ data augment methods such as random crop, multiscale training to augment the dataset. During inference, we employ horizontal flip and softnms techniques to generate more robust resutls.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	1 GPU (Titan X)
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.606804	0.380382	0.226423
cubberly-auditorium-2019-04-22_1	0.52442	0.30373	0.22069
discovery-walk-2019-02-28_0	0.636675	0.458515	0.178161
discovery-walk-2019-02-28_1	0.5838	0.365431	0.218369
food-trucks-2019-02-12_0	0.546014	0.365452	0.180562
gates-ai-lab-2019-04-17_0	0.51494	0.144198	0.370742
gates-basement-elevators-2019-01-17_0	0.510634	0.327934	0.1827
gates-foyer-2019-01-17_0	0.40083	0.150654	0.250177
gates-to-clark-2019-02-28_0	0.449252	0.186914	0.262339
hewlett-class-2019-01-23_0	0.418747	0.163234	0.255513
hewlett-class-2019-01-23_1	0.297585	0.124458	0.173127
huang-2-2019-01-25_1	0.580678	0.344982	0.235696
huang-intersection-2019-01-22_0	0.718987	0.489013	0.229974
indoor-coupa-cafe-2019-02-06_0	0.694714	0.512213	0.182501
lomita-serra-intersection-2019-01-30_0	0.690714	0.49298	0.197735
meyer-green-2019-03-16_1	0.69756	0.406637	0.290922
nvidia-aud-2019-01-25_0	0.60861	0.322143	0.286466
nvidia-aud-2019-04-18_1	0.479979	0.175923	0.304056
nvidia-aud-2019-04-18_2	0.486463	0.200609	0.285854
outdoor-coupa-cafe-2019-02-06_0	0.742562	0.489388	0.253174
quarry-road-2019-02-28_0	0.669985	0.299328	0.370657
serra-street-2019-01-30_0	0.784146	0.612604	0.171542
stlc-111-2019-04-19_1	0.34064	0.123789	0.216851
stlc-111-2019-04-19_2	0.368912	0.12441	0.244502
tressider-2019-03-16_2	0.741607	0.567792	0.173814
tressider-2019-04-26_0	0.73415	0.546872	0.187278
tressider-2019-04-26_1	0.703827	0.567359	0.136468
tressider-2019-04-26_3	0.675858	0.436067	0.239791

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.757656	0.680975	0.451206
cubberly-auditorium-2019-04-22_1	0.810615	0.744256	0.536278
discovery-walk-2019-02-28_0	0.868476	0.829301	0.557611
discovery-walk-2019-02-28_1	0.825303	0.787673	0.574643
food-trucks-2019-02-12_0	0.845253	0.780853	0.602394
gates-ai-lab-2019-04-17_0	0.793074	0.743703	0.628655
gates-basement-elevators-2019-01-17_0	0.900047	0.875985	0.692299
gates-foyer-2019-01-17_0	0.893021	0.871777	0.751938
gates-to-clark-2019-02-28_0	0.829821	0.783403	0.640297
hewlett-class-2019-01-23_0	0.905429	0.896127	0.760618
hewlett-class-2019-01-23_1	0.9474	0.942881	0.866914
huang-2-2019-01-25_1	0.706886	0.644089	0.448664
huang-intersection-2019-01-22_0	0.783	0.7388	0.489541
indoor-coupa-cafe-2019-02-06_0	0.744553	0.650763	0.420619
lomita-serra-intersection-2019-01-30_0	0.764206	0.668129	0.47577
meyer-green-2019-03-16_1	0.636658	0.511225	0.274267
nvidia-aud-2019-01-25_0	0.766555	0.64347	0.335286
nvidia-aud-2019-04-18_1	0.832248	0.791538	0.640583
nvidia-aud-2019-04-18_2	0.869339	0.809435	0.547009
outdoor-coupa-cafe-2019-02-06_0	0.71224	0.58879	0.328154
quarry-road-2019-02-28_0	0.712724	0.66704	0.448802
serra-street-2019-01-30_0	0.738523	0.616872	0.312484
stlc-111-2019-04-19_1	0.871227	0.82651	0.732377
stlc-111-2019-04-19_2	0.849147	0.820208	0.715708
tressider-2019-03-16_2	0.694744	0.61242	0.379305
tressider-2019-04-26_0	0.712096	0.605863	0.337648
tressider-2019-04-26_1	0.74694	0.672534	0.470155
tressider-2019-04-26_3	0.747961	0.646451	0.36395

Team_HJ

0.626

75.075%

67.381%

43.79%

Anonymous Submission

Details

Name	Team_HJ
Submission Date	2021-12-16 22:28:11+00:00
Abstract	Panoramic image detection and tracking using both image and point clouds. For detection, we adopt the Cascade rcnn as our object detector. The backbone of the detector is ResNet 101. The detector is pre-trained on the COCO dataset and then fine tuned on the JRDB dataset. During training, we employ data augment methods such as mixup, random crop, multiscale training to augment the dataset. During inference, we employ horizontal flip and softnms techniques to generate more robust resutls.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.62613	0.427291	0.198839
cubberly-auditorium-2019-04-22_1	0.553812	0.3459	0.207911
discovery-walk-2019-02-28_0	0.647278	0.47502	0.172258
discovery-walk-2019-02-28_1	0.599133	0.394444	0.204688
food-trucks-2019-02-12_0	0.568479	0.408347	0.160132
gates-ai-lab-2019-04-17_0	0.521175	0.169639	0.351536
gates-basement-elevators-2019-01-17_0	0.477859	0.307761	0.170098
gates-foyer-2019-01-17_0	0.397639	0.168638	0.229001
gates-to-clark-2019-02-28_0	0.457175	0.175473	0.281701
hewlett-class-2019-01-23_0	0.432443	0.20714	0.225302
hewlett-class-2019-01-23_1	0.278984	0.106112	0.172872
huang-2-2019-01-25_1	0.610207	0.412843	0.197364
huang-intersection-2019-01-22_0	0.739091	0.556361	0.18273
indoor-coupa-cafe-2019-02-06_0	0.721588	0.583603	0.137985
lomita-serra-intersection-2019-01-30_0	0.706443	0.509923	0.19652
meyer-green-2019-03-16_1	0.729649	0.473158	0.256491
nvidia-aud-2019-01-25_0	0.638617	0.388539	0.250078
nvidia-aud-2019-04-18_1	0.531411	0.362788	0.168623
nvidia-aud-2019-04-18_2	0.514484	0.237079	0.277405
outdoor-coupa-cafe-2019-02-06_0	0.770161	0.56641	0.203751
quarry-road-2019-02-28_0	0.697728	0.352006	0.345722
serra-street-2019-01-30_0	0.793765	0.624009	0.169757
stlc-111-2019-04-19_1	0.35176	0.154387	0.197373
stlc-111-2019-04-19_2	0.37482	0.153421	0.221399
tressider-2019-03-16_2	0.764647	0.614939	0.149708
tressider-2019-04-26_0	0.768734	0.621478	0.147256
tressider-2019-04-26_1	0.731174	0.613661	0.117513
tressider-2019-04-26_3	0.71315	0.521514	0.191637

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.750755	0.673813	0.437903
cubberly-auditorium-2019-04-22_1	0.818955	0.750113	0.504659
discovery-walk-2019-02-28_0	0.886846	0.825894	0.535332
discovery-walk-2019-02-28_1	0.843722	0.78908	0.564708
food-trucks-2019-02-12_0	0.822448	0.770223	0.571592
gates-ai-lab-2019-04-17_0	0.795596	0.743764	0.627432
gates-basement-elevators-2019-01-17_0	0.919045	0.899166	0.655346
gates-foyer-2019-01-17_0	0.906138	0.880852	0.735561
gates-to-clark-2019-02-28_0	0.852584	0.794834	0.643124
hewlett-class-2019-01-23_0	0.893012	0.899794	0.768079
hewlett-class-2019-01-23_1	0.951205	0.94646	0.871291
huang-2-2019-01-25_1	0.692021	0.635411	0.441372
huang-intersection-2019-01-22_0	0.775348	0.731425	0.441324
indoor-coupa-cafe-2019-02-06_0	0.732479	0.658213	0.400321
lomita-serra-intersection-2019-01-30_0	0.778406	0.678567	0.471275
meyer-green-2019-03-16_1	0.619183	0.501898	0.268331
nvidia-aud-2019-01-25_0	0.759362	0.638988	0.339665
nvidia-aud-2019-04-18_1	0.800622	0.741961	0.605697
nvidia-aud-2019-04-18_2	0.873306	0.824491	0.498393
outdoor-coupa-cafe-2019-02-06_0	0.689085	0.564904	0.308967
quarry-road-2019-02-28_0	0.717388	0.672068	0.441882
serra-street-2019-01-30_0	0.769688	0.621595	0.313871
stlc-111-2019-04-19_1	0.886883	0.844966	0.735975
stlc-111-2019-04-19_2	0.887891	0.860709	0.748118
tressider-2019-03-16_2	0.708364	0.635097	0.354738
tressider-2019-04-26_0	0.710497	0.598482	0.323565
tressider-2019-04-26_1	0.730292	0.66692	0.448494
tressider-2019-04-26_3	0.753866	0.647298	0.359689

TEAM_Hojun

0.629

74.219%

65.985%

40.896%

Anonymous Submission

Details

Name	TEAM_Hojun
Submission Date	2021-12-12 21:15:35+00:00
Abstract	State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [1] and Fast R-CNN [2] have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features—using the recently popular terminology of neural networks with “attention” mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model [3], our detection system has a frame rate of 5fps .
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.628799	0.417815	0.210984
cubberly-auditorium-2019-04-22_1	0.562678	0.347934	0.214744
discovery-walk-2019-02-28_0	0.675595	0.515153	0.160442
discovery-walk-2019-02-28_1	0.628072	0.433956	0.194116
food-trucks-2019-02-12_0	0.578375	0.401445	0.176931
gates-ai-lab-2019-04-17_0	0.525692	0.160954	0.364738
gates-basement-elevators-2019-01-17_0	0.516951	0.342662	0.174288
gates-foyer-2019-01-17_0	0.414571	0.171826	0.242745
gates-to-clark-2019-02-28_0	0.471298	0.241063	0.230235
hewlett-class-2019-01-23_0	0.43058	0.191917	0.238663
hewlett-class-2019-01-23_1	0.285055	0.111068	0.173988
huang-2-2019-01-25_1	0.61972	0.41176	0.20796
huang-intersection-2019-01-22_0	0.744511	0.517531	0.22698
indoor-coupa-cafe-2019-02-06_0	0.71052	0.548533	0.161987
lomita-serra-intersection-2019-01-30_0	0.735297	0.566707	0.16859
meyer-green-2019-03-16_1	0.720485	0.467775	0.25271
nvidia-aud-2019-01-25_0	0.63829	0.371537	0.266753
nvidia-aud-2019-04-18_1	0.518909	0.328696	0.190213
nvidia-aud-2019-04-18_2	0.514119	0.262531	0.251588
outdoor-coupa-cafe-2019-02-06_0	0.757459	0.521245	0.236213
quarry-road-2019-02-28_0	0.701909	0.293742	0.408166
serra-street-2019-01-30_0	0.815189	0.675018	0.140171
stlc-111-2019-04-19_1	0.353706	0.136837	0.216869
stlc-111-2019-04-19_2	0.371824	0.133302	0.238522
tressider-2019-03-16_2	0.766779	0.615002	0.151778
tressider-2019-04-26_0	0.754943	0.58553	0.169413
tressider-2019-04-26_1	0.715559	0.578154	0.137405
tressider-2019-04-26_3	0.685805	0.453583	0.232222

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.742188	0.659853	0.408959
cubberly-auditorium-2019-04-22_1	0.773547	0.70433	0.440094
discovery-walk-2019-02-28_0	0.838622	0.762332	0.492556
discovery-walk-2019-02-28_1	0.773286	0.7228	0.507019
food-trucks-2019-02-12_0	0.807738	0.753931	0.536014
gates-ai-lab-2019-04-17_0	0.793145	0.73457	0.597822
gates-basement-elevators-2019-01-17_0	0.906171	0.882132	0.711729
gates-foyer-2019-01-17_0	0.8965	0.876479	0.726237
gates-to-clark-2019-02-28_0	0.801089	0.753789	0.619459
hewlett-class-2019-01-23_0	0.905675	0.891039	0.732253
hewlett-class-2019-01-23_1	0.950136	0.944583	0.863834
huang-2-2019-01-25_1	0.636321	0.575494	0.392757
huang-intersection-2019-01-22_0	0.729623	0.6802	0.423228
indoor-coupa-cafe-2019-02-06_0	0.721582	0.629018	0.401261
lomita-serra-intersection-2019-01-30_0	0.704412	0.628361	0.432809
meyer-green-2019-03-16_1	0.603607	0.483064	0.247971
nvidia-aud-2019-01-25_0	0.724831	0.603566	0.298653
nvidia-aud-2019-04-18_1	0.792987	0.754671	0.611029
nvidia-aud-2019-04-18_2	0.874401	0.799069	0.447102
outdoor-coupa-cafe-2019-02-06_0	0.690512	0.543669	0.286
quarry-road-2019-02-28_0	0.641651	0.599787	0.382726
serra-street-2019-01-30_0	0.70999	0.579609	0.270406
stlc-111-2019-04-19_1	0.871005	0.825056	0.716747
stlc-111-2019-04-19_2	0.85055	0.806347	0.691534
tressider-2019-03-16_2	0.687199	0.590007	0.300789
tressider-2019-04-26_0	0.706384	0.577162	0.309167
tressider-2019-04-26_1	0.747083	0.662719	0.418368
tressider-2019-04-26_3	0.751483	0.636808	0.319124

TEST_KKANG

0.668

65.424%

59.718%

40.158%

Anonymous Submission

Details

Name	TEST_KKANG
Submission Date	2021-12-12 21:17:24+00:00
Abstract	State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features---using the recently popular terminology of neural networks with attention mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model, our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.668004	0.521231	0.146773
cubberly-auditorium-2019-04-22_1	0.590781	0.409059	0.181721
discovery-walk-2019-02-28_0	0.6372	0.447629	0.189571
discovery-walk-2019-02-28_1	0.6016	0.407586	0.194014
food-trucks-2019-02-12_0	0.606389	0.474102	0.132287
gates-ai-lab-2019-04-17_0	0.586345	0.467421	0.118924
gates-basement-elevators-2019-01-17_0	0.57726	0.465785	0.111475
gates-foyer-2019-01-17_0	0.446727	0.290253	0.156475
gates-to-clark-2019-02-28_0	0.459744	0.269143	0.190601
hewlett-class-2019-01-23_0	0.518349	0.38887	0.12948
hewlett-class-2019-01-23_1	0.288706	0.13674	0.151965
huang-2-2019-01-25_1	0.665563	0.516824	0.148739
huang-intersection-2019-01-22_0	0.736653	0.551087	0.185566
indoor-coupa-cafe-2019-02-06_0	0.789486	0.706129	0.083357
lomita-serra-intersection-2019-01-30_0	0.713124	0.467106	0.246018
meyer-green-2019-03-16_1	0.747229	0.538988	0.208241
nvidia-aud-2019-01-25_0	0.75395	0.638882	0.115068
nvidia-aud-2019-04-18_1	0.521184	0.397236	0.123948
nvidia-aud-2019-04-18_2	0.589375	0.431127	0.158248
outdoor-coupa-cafe-2019-02-06_0	0.814607	0.689223	0.125384
quarry-road-2019-02-28_0	0.680732	0.302053	0.378678
serra-street-2019-01-30_0	0.79694	0.632355	0.164585
stlc-111-2019-04-19_1	0.374219	0.188627	0.185592
stlc-111-2019-04-19_2	0.404019	0.194239	0.20978
tressider-2019-03-16_2	0.766553	0.644948	0.121605
tressider-2019-04-26_0	0.847977	0.774191	0.073785
tressider-2019-04-26_1	0.784969	0.704245	0.080725
tressider-2019-04-26_3	0.796712	0.691583	0.10513

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.654237	0.597179	0.401582
cubberly-auditorium-2019-04-22_1	0.755804	0.686425	0.45734
discovery-walk-2019-02-28_0	0.87488	0.82107	0.50096
discovery-walk-2019-02-28_1	0.811319	0.776177	0.556669
food-trucks-2019-02-12_0	0.754846	0.705726	0.533179
gates-ai-lab-2019-04-17_0	0.748501	0.721874	0.613902
gates-basement-elevators-2019-01-17_0	0.833774	0.828847	0.626927
gates-foyer-2019-01-17_0	0.831803	0.805291	0.686808
gates-to-clark-2019-02-28_0	0.814754	0.764886	0.618913
hewlett-class-2019-01-23_0	0.854477	0.855064	0.711151
hewlett-class-2019-01-23_1	0.914789	0.913025	0.831072
huang-2-2019-01-25_1	0.594004	0.544261	0.414868
huang-intersection-2019-01-22_0	0.787538	0.720659	0.418163
indoor-coupa-cafe-2019-02-06_0	0.598521	0.54467	0.389395
lomita-serra-intersection-2019-01-30_0	0.785522	0.678122	0.397123
meyer-green-2019-03-16_1	0.578104	0.491044	0.255725
nvidia-aud-2019-01-25_0	0.587233	0.503495	0.3108
nvidia-aud-2019-04-18_1	0.775983	0.756122	0.651423
nvidia-aud-2019-04-18_2	0.765037	0.708592	0.449966
outdoor-coupa-cafe-2019-02-06_0	0.600024	0.503968	0.30511
quarry-road-2019-02-28_0	0.719371	0.673935	0.413468
serra-street-2019-01-30_0	0.740402	0.603288	0.286938
stlc-111-2019-04-19_1	0.853176	0.805537	0.686435
stlc-111-2019-04-19_2	0.847729	0.819135	0.711614
tressider-2019-03-16_2	0.699591	0.61979	0.320554
tressider-2019-04-26_0	0.51013	0.449797	0.28122
tressider-2019-04-26_1	0.604933	0.568503	0.40069
tressider-2019-04-26_3	0.618181	0.532633	0.31578

Faster R-CNN

0.682

61.787%

52.168%

29.515%

S. Ren, K. He, R. Girshick and J. Sun Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. in NeurIPS, 2015

Details

Name	Faster R-CNN
Submission Date	2021-12-16 22:33:02+00:00
Abstract	State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features---using the recently popular terminology of neural networks with attention mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model, our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.
Publication title	Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.	Publication authors:	S. Ren, K. He, R. Girshick and J. Sun
Publication venue and year	NeurIPS, 2015	Publication URL:	https://arxiv.org/abs/1506.01497
Code Language	N/A	Hardware	1 GPU (Titan X)
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.682367	0.407706	0.27466
cubberly-auditorium-2019-04-22_1	0.620242	0.28442	0.335822
discovery-walk-2019-02-28_0	0.680303	0.389538	0.290765
discovery-walk-2019-02-28_1	0.636489	0.320349	0.316139
food-trucks-2019-02-12_0	0.678212	0.532864	0.145348
gates-ai-lab-2019-04-17_0	0.644044	0.243731	0.400312
gates-basement-elevators-2019-01-17_0	0.54184	0.30795	0.233891
gates-foyer-2019-01-17_0	0.557313	0.261761	0.295552
gates-to-clark-2019-02-28_0	0.508497	0.201777	0.30672
hewlett-class-2019-01-23_0	0.46913	0.245667	0.223463
hewlett-class-2019-01-23_1	0.325886	0.12125	0.204636
huang-2-2019-01-25_1	0.620261	0.182669	0.437592
huang-intersection-2019-01-22_0	0.772304	0.421952	0.350352
indoor-coupa-cafe-2019-02-06_0	0.749721	0.531798	0.217922
lomita-serra-intersection-2019-01-30_0	0.763183	0.47194	0.291243
meyer-green-2019-03-16_1	0.790705	0.526574	0.264131
nvidia-aud-2019-01-25_0	0.690111	0.336769	0.353342
nvidia-aud-2019-04-18_1	0.551259	0.229284	0.321974
nvidia-aud-2019-04-18_2	0.625954	0.176389	0.449565
outdoor-coupa-cafe-2019-02-06_0	0.827472	0.602667	0.224805
quarry-road-2019-02-28_0	0.762108	0.222416	0.539692
serra-street-2019-01-30_0	0.829953	0.521857	0.308096
stlc-111-2019-04-19_1	0.408206	0.176219	0.231988
stlc-111-2019-04-19_2	0.449293	0.158268	0.291025
tressider-2019-03-16_2	0.776865	0.480777	0.296088
tressider-2019-04-26_0	0.801941	0.64258	0.159361
tressider-2019-04-26_1	0.745903	0.596439	0.149464
tressider-2019-04-26_3	0.73506	0.492982	0.242077

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.617872	0.521683	0.295153
cubberly-auditorium-2019-04-22_1	0.665964	0.538905	0.298307
discovery-walk-2019-02-28_0	0.753944	0.646111	0.370385
discovery-walk-2019-02-28_1	0.701051	0.639289	0.434888
food-trucks-2019-02-12_0	0.615683	0.566419	0.384295
gates-ai-lab-2019-04-17_0	0.707694	0.655841	0.510198
gates-basement-elevators-2019-01-17_0	0.804578	0.778151	0.539282
gates-foyer-2019-01-17_0	0.845786	0.813596	0.631383
gates-to-clark-2019-02-28_0	0.759119	0.693947	0.546687
hewlett-class-2019-01-23_0	0.882362	0.829594	0.659174
hewlett-class-2019-01-23_1	0.945475	0.921831	0.740138
huang-2-2019-01-25_1	0.628159	0.537175	0.316693
huang-intersection-2019-01-22_0	0.647857	0.546635	0.256462
indoor-coupa-cafe-2019-02-06_0	0.607555	0.506504	0.260568
lomita-serra-intersection-2019-01-30_0	0.552051	0.48707	0.275702
meyer-green-2019-03-16_1	0.40193	0.300718	0.147217
nvidia-aud-2019-01-25_0	0.613843	0.427921	0.199674
nvidia-aud-2019-04-18_1	0.765102	0.596392	0.4737
nvidia-aud-2019-04-18_2	0.756504	0.612252	0.313054
outdoor-coupa-cafe-2019-02-06_0	0.517552	0.366091	0.163145
quarry-road-2019-02-28_0	0.551752	0.492795	0.278364
serra-street-2019-01-30_0	0.53812	0.434496	0.210049
stlc-111-2019-04-19_1	0.814259	0.764297	0.608936
stlc-111-2019-04-19_2	0.790533	0.746317	0.598733
tressider-2019-03-16_2	0.524153	0.432897	0.256363
tressider-2019-04-26_0	0.547068	0.434685	0.197159
tressider-2019-04-26_1	0.629784	0.565671	0.320058
tressider-2019-04-26_3	0.668653	0.524041	0.220059

DETR

0.721

63.884%

48.664%

22.769%

N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, S. Zagoruyko. End-to-End Object Detection with Transformers. in ECCV, 2020

Details

Name	DETR
Submission Date	2021-12-12 21:19:43+00:00
Abstract	We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries, DETR reasons about the relations of the objects and the global image context to directly output the final set of predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and highly-optimized Faster R-CNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive baselines. Training code and pretrained models are available at https://github.com/facebookresearch/detr.
Publication title	End-to-End Object Detection with Transformers.	Publication authors:	N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, S. Zagoruyko.
Publication venue and year	ECCV, 2020	Publication URL:	https://arxiv.org/abs/2005.12872
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.721049	0.35819	0.362858
cubberly-auditorium-2019-04-22_1	0.620212	0.286367	0.333846
discovery-walk-2019-02-28_0	0.803924	0.542164	0.26176
discovery-walk-2019-02-28_1	0.824686	0.604289	0.220397
food-trucks-2019-02-12_0	0.667527	0.296671	0.370856
gates-ai-lab-2019-04-17_0	0.693475	0.401032	0.292443
gates-basement-elevators-2019-01-17_0	0.647143	0.3087	0.338443
gates-foyer-2019-01-17_0	0.663322	0.4439	0.219422
gates-to-clark-2019-02-28_0	0.767025	0.604577	0.162447
hewlett-class-2019-01-23_0	0.620925	0.377578	0.243347
hewlett-class-2019-01-23_1	0.536491	0.357903	0.178588
huang-2-2019-01-25_1	0.670552	0.237298	0.433254
huang-intersection-2019-01-22_0	0.845283	0.518615	0.326667
indoor-coupa-cafe-2019-02-06_0	0.700435	0.186747	0.513688
lomita-serra-intersection-2019-01-30_0	0.874222	0.6437	0.230521
meyer-green-2019-03-16_1	0.827806	0.463307	0.364499
nvidia-aud-2019-01-25_0	0.71697	0.246864	0.470106
nvidia-aud-2019-04-18_1	0.668983	0.272949	0.396035
nvidia-aud-2019-04-18_2	0.720635	0.434083	0.286552
outdoor-coupa-cafe-2019-02-06_0	0.792229	0.311535	0.480695
quarry-road-2019-02-28_0	0.90995	0.762005	0.147945
serra-street-2019-01-30_0	0.843765	0.421787	0.421978
stlc-111-2019-04-19_1	0.518518	0.280392	0.238126
stlc-111-2019-04-19_2	0.529957	0.251425	0.278532
tressider-2019-03-16_2	0.810511	0.410176	0.400335
tressider-2019-04-26_0	0.716928	0.253696	0.463232
tressider-2019-04-26_1	0.664791	0.23657	0.428222
tressider-2019-04-26_3	0.65932	0.089598	0.569722

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.63884	0.48664	0.227687
cubberly-auditorium-2019-04-22_1	0.723547	0.614066	0.372413
discovery-walk-2019-02-28_0	0.700447	0.549403	0.33463
discovery-walk-2019-02-28_1	0.680343	0.576734	0.32348
food-trucks-2019-02-12_0	0.731617	0.597451	0.296803
gates-ai-lab-2019-04-17_0	0.757919	0.684306	0.510542
gates-basement-elevators-2019-01-17_0	0.760553	0.690566	0.466285
gates-foyer-2019-01-17_0	0.811411	0.747334	0.520463
gates-to-clark-2019-02-28_0	0.741837	0.636902	0.439991
hewlett-class-2019-01-23_0	0.828101	0.788844	0.561356
hewlett-class-2019-01-23_1	0.913579	0.885695	0.669948
huang-2-2019-01-25_1	0.610156	0.497532	0.270375
huang-intersection-2019-01-22_0	0.651404	0.456637	0.179137
indoor-coupa-cafe-2019-02-06_0	0.620138	0.478063	0.233034
lomita-serra-intersection-2019-01-30_0	0.603811	0.383418	0.235862
meyer-green-2019-03-16_1	0.476046	0.313803	0.130892
nvidia-aud-2019-01-25_0	0.577296	0.416606	0.15637
nvidia-aud-2019-04-18_1	0.673762	0.604265	0.389516
nvidia-aud-2019-04-18_2	0.754536	0.630602	0.141214
outdoor-coupa-cafe-2019-02-06_0	0.570202	0.397808	0.168729
quarry-road-2019-02-28_0	0.537423	0.413082	0.216594
serra-street-2019-01-30_0	0.587069	0.358779	0.117521
stlc-111-2019-04-19_1	0.823997	0.759333	0.613439
stlc-111-2019-04-19_2	0.828125	0.742592	0.513629
tressider-2019-03-16_2	0.585544	0.458202	0.157857
tressider-2019-04-26_0	0.569181	0.384417	0.145783
tressider-2019-04-26_1	0.627949	0.476278	0.20106
tressider-2019-04-26_3	0.621362	0.423914	0.160175

RetinaNet

0.733

61.861%

50.38%

27.165%

T. Lin, P. Goyal, R. Girshick, K. He and P. Dollár. Focal Loss for Dense Object Detection in ICCV, 2017

Details

Name	RetinaNet
Submission Date	2021-12-12 21:22:04+00:00
Abstract	The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors.
Publication title	Focal Loss for Dense Object Detection	Publication authors:	T. Lin, P. Goyal, R. Girshick, K. He and P. Dollár.
Publication venue and year	ICCV, 2017	Publication URL:	https://arxiv.org/abs/1708.02002
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	https://github.com/facebookresearch/detectron2

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.733039	0.476553	0.256486
cubberly-auditorium-2019-04-22_1	0.686208	0.420819	0.265389
discovery-walk-2019-02-28_0	0.771539	0.43042	0.341119
discovery-walk-2019-02-28_1	0.725696	0.418803	0.306892
food-trucks-2019-02-12_0	0.745576	0.57769	0.167886
gates-ai-lab-2019-04-17_0	0.615517	0.126098	0.489419
gates-basement-elevators-2019-01-17_0	0.626486	0.428004	0.198482
gates-foyer-2019-01-17_0	0.517044	0.178179	0.338865
gates-to-clark-2019-02-28_0	0.59681	0.400176	0.196633
hewlett-class-2019-01-23_0	0.506483	0.218096	0.288387
hewlett-class-2019-01-23_1	0.400005	0.141885	0.25812
huang-2-2019-01-25_1	0.697224	0.447644	0.24958
huang-intersection-2019-01-22_0	0.835321	0.507061	0.328259
indoor-coupa-cafe-2019-02-06_0	0.78696	0.570463	0.216498
lomita-serra-intersection-2019-01-30_0	0.83888	0.660893	0.177987
meyer-green-2019-03-16_1	0.846384	0.630049	0.216335
nvidia-aud-2019-01-25_0	0.764505	0.493568	0.270938
nvidia-aud-2019-04-18_1	0.602248	0.355329	0.246919
nvidia-aud-2019-04-18_2	0.665507	0.215727	0.449779
outdoor-coupa-cafe-2019-02-06_0	0.877153	0.71638	0.160773
quarry-road-2019-02-28_0	0.83197	0.443431	0.388539
serra-street-2019-01-30_0	0.89945	0.660625	0.238825
stlc-111-2019-04-19_1	0.466651	0.225915	0.240737
stlc-111-2019-04-19_2	0.531633	0.297411	0.234222
tressider-2019-03-16_2	0.864894	0.642972	0.221922
tressider-2019-04-26_0	0.83312	0.658001	0.175119
tressider-2019-04-26_1	0.784863	0.594492	0.190371
tressider-2019-04-26_3	0.783221	0.528582	0.254639

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.618613	0.503799	0.271654
cubberly-auditorium-2019-04-22_1	0.698412	0.532919	0.303489
discovery-walk-2019-02-28_0	0.637999	0.482166	0.319312
discovery-walk-2019-02-28_1	0.655071	0.562439	0.377051
food-trucks-2019-02-12_0	0.648788	0.559675	0.29741
gates-ai-lab-2019-04-17_0	0.723351	0.668771	0.496505
gates-basement-elevators-2019-01-17_0	0.824934	0.765753	0.486265
gates-foyer-2019-01-17_0	0.819324	0.798636	0.643025
gates-to-clark-2019-02-28_0	0.682696	0.626826	0.513614
hewlett-class-2019-01-23_0	0.896748	0.867171	0.638641
hewlett-class-2019-01-23_1	0.912327	0.874166	0.741679
huang-2-2019-01-25_1	0.603457	0.482307	0.302362
huang-intersection-2019-01-22_0	0.563418	0.455224	0.246814
indoor-coupa-cafe-2019-02-06_0	0.640923	0.509853	0.25657
lomita-serra-intersection-2019-01-30_0	0.503392	0.420568	0.236863
meyer-green-2019-03-16_1	0.436563	0.308499	0.154023
nvidia-aud-2019-01-25_0	0.590848	0.425777	0.195973
nvidia-aud-2019-04-18_1	0.782246	0.612574	0.443556
nvidia-aud-2019-04-18_2	0.668533	0.557414	0.276856
outdoor-coupa-cafe-2019-02-06_0	0.498281	0.34839	0.14933
quarry-road-2019-02-28_0	0.458454	0.398908	0.264411
serra-street-2019-01-30_0	0.412998	0.321111	0.170244
stlc-111-2019-04-19_1	0.820841	0.755117	0.608457
stlc-111-2019-04-19_2	0.78616	0.733923	0.598604
tressider-2019-03-16_2	0.48619	0.366394	0.166229
tressider-2019-04-26_0	0.590539	0.433025	0.194134
tressider-2019-04-26_1	0.634749	0.529644	0.293251
tressider-2019-04-26_3	0.672648	0.498012	0.185408

YOLOv3

0.74

55.569%

41.731%

20.906%

J. Redmon and A. Farhadi YOLOv3: An Incremental Improvement in arXiv, 2018

Details

Name	YOLOv3
Submission Date	2021-12-12 21:23:46+00:00
Abstract	We present some updates to YOLO! We made a bunch of little design changes to make it better. We also trained this new network that’s pretty swell. It’s a little bigger than last time but more accurate. It’s still fast though, don’t worry. At 320 × 320 YOLOv3 runs in 22 ms at 28.2 mAP, as accurate as SSD but three times faster. When we look at the old .5 IOU mAP detection metric YOLOv3 is quite good. It achieves 57.9 AP50 in 51 ms on a Titan X, compared to 57.5 AP50 in 198 ms by RetinaNet, similar performance but 3.8× faster.
Publication title	YOLOv3: An Incremental Improvement	Publication authors:	J. Redmon and A. Farhadi
Publication venue and year	arXiv, 2018	Publication URL:	https://arxiv.org/abs/1804.02767
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

hewlett-class-2019-01-23_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.740138	0.558715	0.181422
cubberly-auditorium-2019-04-22_1	0.723683	0.558119	0.165564
discovery-walk-2019-02-28_0	0.712232	0.420013	0.292219
discovery-walk-2019-02-28_1	0.713975	0.486056	0.227919
food-trucks-2019-02-12_0	0.699629	0.538973	0.160655
gates-ai-lab-2019-04-17_0	0.635971	0.451487	0.184484
gates-basement-elevators-2019-01-17_0	0.714368	0.502261	0.212107
gates-foyer-2019-01-17_0	0.499795	0.262883	0.236913
gates-to-clark-2019-02-28_0	0.529793	0.30667	0.223122
hewlett-class-2019-01-23_0	0.580017	0.336978	0.243039
hewlett-class-2019-01-23_1	0.407532	0.157862	0.24967
huang-2-2019-01-25_1	0.734212	0.501751	0.232461
huang-intersection-2019-01-22_0	0.808658	0.653928	0.15473
indoor-coupa-cafe-2019-02-06_0	0.834673	0.703847	0.130827
lomita-serra-intersection-2019-01-30_0	0.776282	0.576813	0.199469
meyer-green-2019-03-16_1	0.842208	0.729059	0.113149
nvidia-aud-2019-01-25_0	0.805178	0.618923	0.186255
nvidia-aud-2019-04-18_1	0.629439	0.442768	0.186671
nvidia-aud-2019-04-18_2	0.765908	0.499462	0.266446
outdoor-coupa-cafe-2019-02-06_0	0.88881	0.795499	0.093311
quarry-road-2019-02-28_0	0.772072	0.543698	0.228373
serra-street-2019-01-30_0	0.817898	0.524226	0.293672
stlc-111-2019-04-19_1	0.466803	0.204872	0.261931
stlc-111-2019-04-19_2	0.510321	0.200215	0.310106
tressider-2019-03-16_2	0.856154	0.76294	0.093214
tressider-2019-04-26_0	0.875264	0.754111	0.121152
tressider-2019-04-26_1	0.834457	0.727695	0.106762
tressider-2019-04-26_3	0.8458	0.714986	0.130815

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.555693	0.417308	0.209057
cubberly-auditorium-2019-04-22_1	0.62883	0.52121	0.26164
discovery-walk-2019-02-28_0	0.721027	0.542249	0.27816
discovery-walk-2019-02-28_1	0.641029	0.526636	0.275804
food-trucks-2019-02-12_0	0.651749	0.589604	0.310848
gates-ai-lab-2019-04-17_0	0.698891	0.620315	0.423431
gates-basement-elevators-2019-01-17_0	0.699636	0.635444	0.314457
gates-foyer-2019-01-17_0	0.798834	0.730365	0.505041
gates-to-clark-2019-02-28_0	0.762953	0.681083	0.467156
hewlett-class-2019-01-23_0	0.768261	0.724606	0.421673
hewlett-class-2019-01-23_1	0.899454	0.819074	0.604506
huang-2-2019-01-25_1	0.488022	0.338966	0.203026
huang-intersection-2019-01-22_0	0.644776	0.516353	0.192228
indoor-coupa-cafe-2019-02-06_0	0.531393	0.37629	0.197861
lomita-serra-intersection-2019-01-30_0	0.626843	0.500425	0.222101
meyer-green-2019-03-16_1	0.460875	0.320416	0.143145
nvidia-aud-2019-01-25_0	0.505603	0.327894	0.15638
nvidia-aud-2019-04-18_1	0.722888	0.646028	0.49011
nvidia-aud-2019-04-18_2	0.557115	0.303314	0.0963535
outdoor-coupa-cafe-2019-02-06_0	0.488643	0.323565	0.140062
quarry-road-2019-02-28_0	0.579021	0.496468	0.213296
serra-street-2019-01-30_0	0.526019	0.326028	0.111461
stlc-111-2019-04-19_1	0.771185	0.707106	0.454402
stlc-111-2019-04-19_2	0.756436	0.645326	0.436698
tressider-2019-03-16_2	0.513752	0.418646	0.172916
tressider-2019-04-26_0	0.470524	0.263253	0.112707
tressider-2019-04-26_1	0.494651	0.3811	0.19316
tressider-2019-04-26_3	0.506128	0.348149	0.15529

jihoo_S2

0.957

2.031%

0.365%

0.022%

Anonymous Submission

Details

Name	jihoo_S2
Submission Date	2021-12-12 21:24:34+00:00
Abstract	State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features---using the recently popular terminology of neural networks with attention mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model, our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

tressider-2019-04-26_1

Best 3 By AP_0.5 ↓

Worst 3 By AP_0.5 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.95687	0.541449	0.41542
cubberly-auditorium-2019-04-22_1	0.961117	0.400317	0.5608
discovery-walk-2019-02-28_0	0.970517	0.398652	0.571865
discovery-walk-2019-02-28_1	0.963299	0.421712	0.541587
food-trucks-2019-02-12_0	0.936791	0.592035	0.344755
gates-ai-lab-2019-04-17_0	0.938342	0.429964	0.508379
gates-basement-elevators-2019-01-17_0	0.957787	0.424115	0.533672
gates-foyer-2019-01-17_0	0.96554	0.510204	0.455335
gates-to-clark-2019-02-28_0	0.968792	0.509164	0.459628
hewlett-class-2019-01-23_0	0.928409	0.436566	0.491843
hewlett-class-2019-01-23_1	0.954577	0.499889	0.454688
huang-2-2019-01-25_1	0.962189	0.472603	0.489586
huang-intersection-2019-01-22_0	0.972935	0.498437	0.474497
indoor-coupa-cafe-2019-02-06_0	0.939911	0.613065	0.326846
lomita-serra-intersection-2019-01-30_0	0.971055	0.389943	0.581111
meyer-green-2019-03-16_1	0.967464	0.460847	0.506617
nvidia-aud-2019-01-25_0	0.948983	0.50144	0.447543
nvidia-aud-2019-04-18_1	0.954986	0.386153	0.568833
nvidia-aud-2019-04-18_2	0.941554	0.417657	0.523897
outdoor-coupa-cafe-2019-02-06_0	0.960457	0.547928	0.412529
quarry-road-2019-02-28_0	0.969796	0.360467	0.609328
serra-street-2019-01-30_0	0.97792	0.54115	0.43677
stlc-111-2019-04-19_1	0.9363	0.363702	0.572598
stlc-111-2019-04-19_2	0.956092	0.464163	0.491929
tressider-2019-03-16_2	0.969906	0.614441	0.355464
tressider-2019-04-26_0	0.960068	0.834535	0.125534
tressider-2019-04-26_1	0.954	0.853816	0.100184
tressider-2019-04-26_3	0.951631	0.763961	0.18767

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.0203125	0.00365395	0.000218687
cubberly-auditorium-2019-04-22_1	0.00909763	0.00152272	9.5209e-05
discovery-walk-2019-02-28_0	0.00569566	0.000712388	7.67667e-05
discovery-walk-2019-02-28_1	0.00817559	0.00168365	9.15168e-05
food-trucks-2019-02-12_0	0.0361972	0.00974608	0.000576039
gates-ai-lab-2019-04-17_0	0.0250463	0.00382857	0.000184458
gates-basement-elevators-2019-01-17_0	0.00934186	0.00120816	0.000120715
gates-foyer-2019-01-17_0	0.00871875	0.00108774	5.83349e-05
gates-to-clark-2019-02-28_0	0.00601146	0.00092901	4.65513e-05
hewlett-class-2019-01-23_0	0.0333972	0.00678151	0.000265219
hewlett-class-2019-01-23_1	0.0177243	0.00292578	9.36444e-05
huang-2-2019-01-25_1	0.00931908	0.00140609	5.72115e-05
huang-intersection-2019-01-22_0	0.00733661	0.00108755	7.65069e-05
indoor-coupa-cafe-2019-02-06_0	0.0362902	0.00448701	0.000207693
lomita-serra-intersection-2019-01-30_0	0.00509446	0.000686939	6.70122e-05
meyer-green-2019-03-16_1	0.00597816	0.000784453	6.2238e-05
nvidia-aud-2019-01-25_0	0.0194164	0.0017486	0.000121369
nvidia-aud-2019-04-18_1	0.0102203	0.00141234	9.41324e-05
nvidia-aud-2019-04-18_2	0.0177646	0.00297734	0.00016357
outdoor-coupa-cafe-2019-02-06_0	0.0114421	0.00127063	7.20762e-05
quarry-road-2019-02-28_0	0.00462483	0.000565467	5.68182e-05
serra-street-2019-01-30_0	0.00417547	0.000555215	3.77962e-05
stlc-111-2019-04-19_1	0.0208253	0.00310344	0.000166168
stlc-111-2019-04-19_2	0.0111523	0.00149281	7.9324e-05
tressider-2019-03-16_2	0.00802653	0.00135631	0.00014703
tressider-2019-04-26_0	0.0507656	0.00573495	0.000347658
tressider-2019-04-26_1	0.0706468	0.0155778	0.00107682
tressider-2019-04-26_3	0.0435963	0.00765164	0.000412789

3D Detection Submissions

Name

OSPA_IoU↓

AP_0.3 ↑

AP_0.5 ↑

AP_0.7 ↑

DCCLA

0.557

76.282%

47.436%

7.053%

Anonymous Submission

Details

Name	DCCLA
Submission Date	2023-11-23 17:44:57+00:00
Abstract	DCCLA: Dense Cross Connections with Linear Attention for LiDAR-based 3D Pedestrian Detection DCCLA: Dense Cross Connections with Linear Attention for LiDAR-based 3D Pedestrian Detection DCCLA: Dense Cross Connections with Linear Attention for LiDAR-based 3D Pedestrian Detection
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

Generating discovery-walk-2019-02-28_0, Come Back Later!

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.556771	0.21453	0.342241
cubberly-auditorium-2019-04-22_1	0.537779	0.193763	0.344016
discovery-walk-2019-02-28_0	0.608214	0.306565	0.30165
discovery-walk-2019-02-28_1	0.60775	0.281887	0.325863
food-trucks-2019-02-12_0	0.533731	0.182849	0.350882
gates-ai-lab-2019-04-17_0	0.587568	0.275189	0.312379
gates-basement-elevators-2019-01-17_0	0.457233	0.098194	0.359039
gates-foyer-2019-01-17_0	0.474141	0.161898	0.312243
gates-to-clark-2019-02-28_0	0.4755	0.143317	0.332183
hewlett-class-2019-01-23_0	0.411917	0.122767	0.28915
hewlett-class-2019-01-23_1	0.360728	0.062821	0.297906
huang-2-2019-01-25_1	0.546013	0.219756	0.326258
huang-intersection-2019-01-22_0	0.701667	0.437031	0.264636
indoor-coupa-cafe-2019-02-06_0	0.582132	0.207947	0.374185
lomita-serra-intersection-2019-01-30_0	0.645443	0.247735	0.397708
meyer-green-2019-03-16_1	0.707961	0.346861	0.3611
nvidia-aud-2019-01-25_0	0.636198	0.260101	0.376097
nvidia-aud-2019-04-18_1	0.51355	0.137542	0.376008
nvidia-aud-2019-04-18_2	0.528376	0.10363	0.424747
outdoor-coupa-cafe-2019-02-06_0	0.692399	0.331887	0.360512
quarry-road-2019-02-28_0	0.620436	0.172767	0.447668
serra-street-2019-01-30_0	0.743526	0.504361	0.239165
stlc-111-2019-04-19_1	0.369895	0.086449	0.283446
stlc-111-2019-04-19_2	0.419691	0.111571	0.308119
tressider-2019-03-16_2	0.669169	0.422795	0.246375
tressider-2019-04-26_0	0.571121	0.171736	0.399385
tressider-2019-04-26_1	0.490373	0.088759	0.401614
tressider-2019-04-26_3	0.540297	0.112136	0.428162

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.762819	0.474365	0.0705346
cubberly-auditorium-2019-04-22_1	0.826515	0.463466	0.0530925
discovery-walk-2019-02-28_0	0.962242	0.686212	0.164899
discovery-walk-2019-02-28_1	0.879201	0.541437	0.0668884
food-trucks-2019-02-12_0	0.848253	0.556007	0.0816429
gates-ai-lab-2019-04-17_0	0.767642	0.603537	0.126797
gates-basement-elevators-2019-01-17_0	0.873893	0.581944	0.131367
gates-foyer-2019-01-17_0	0.90705	0.739625	0.206227
gates-to-clark-2019-02-28_0	0.939348	0.737775	0.262978
hewlett-class-2019-01-23_0	0.869322	0.688753	0.199071
hewlett-class-2019-01-23_1	0.961532	0.827261	0.303311
huang-2-2019-01-25_1	0.806333	0.592794	0.13625
huang-intersection-2019-01-22_0	0.898139	0.565653	0.0694548
indoor-coupa-cafe-2019-02-06_0	0.629847	0.39084	0.0482052
lomita-serra-intersection-2019-01-30_0	0.946733	0.743747	0.133514
meyer-green-2019-03-16_1	0.661131	0.371148	0.0882186
nvidia-aud-2019-01-25_0	0.658812	0.378337	0.0626193
nvidia-aud-2019-04-18_1	0.798384	0.572357	0.148207
nvidia-aud-2019-04-18_2	0.841859	0.568647	0.114163
outdoor-coupa-cafe-2019-02-06_0	0.638985	0.297332	0.0242304
quarry-road-2019-02-28_0	0.855253	0.68299	0.202141
serra-street-2019-01-30_0	0.914252	0.540453	0.0917586
stlc-111-2019-04-19_1	0.943594	0.838511	0.421655
stlc-111-2019-04-19_2	0.879517	0.776766	0.32139
tressider-2019-03-16_2	0.850271	0.620468	0.0707458
tressider-2019-04-26_0	0.700118	0.368587	0.0436788
tressider-2019-04-26_1	0.789357	0.505749	0.0418275
tressider-2019-04-26_3	0.713602	0.361807	0.0415111

RPEA

0.572

76.905%

46.076%

5.296%

Jinzheng Guang, Zhengxi Hu, Shichao Wu, Qianyi Zhang, Jingtai Liu∗ RPEA: A Residual Path Network with Efficient Attention for 3D Pedestrian Detection from Point Clouds in Expert Systems With Applications, 2024

Details

Name	RPEA
Submission Date	2023-07-04 11:51:47+00:00
Abstract	Efficiently detecting pedestrians from 3D point cloud data is a significantly challenging perception task in numerous robotic and autonomous driving applications, primarily because of the sparsity of point cloud data representing pedestrian objects and the significant deformations in pedestrian body poses. To address these challenges, we present a Residual Path network with Efficient Attention (RPEA), an end-to-end trainable single-stage 3D pedestrian detection network. We first introduce the Residual Path (ResPath) architecture, which incorporates multiple residual blocks to retain the spatial information lost during downsampling and combines features of different resolution scales. To suppress noise in point clouds while generalizing various pedestrian representations, we propose an efficient Channel Attention module with Average and Maximum pooling strategies (CAAM). Experimental results demonstrate that our RPEA ranks first on the JRDB 3D object detection leaderboard among all solutions, while significantly surpassing the ranks second by up to 5.6% average precision. Additionally, our RPEA achieves real-time pedestrian detection at 39 frame-per-second (fps). Since our method has higher accuracy and faster inference, it can be deployed more effectively in vehicles and mobile robots. The code is publicly available at https://github.com/jinzhengguang/RPEA.
Publication title	RPEA: A Residual Path Network with Efficient Attention for 3D Pedestrian Detection from Point Clouds	Publication authors:	Jinzheng Guang, Zhengxi Hu, Shichao Wu, Qianyi Zhang, Jingtai Liu∗
Publication venue and year	Expert Systems With Applications, 2024	Publication URL:	https://doi.org/10.1016/j.eswa.2024.123497
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

Generating hewlett-class-2019-01-23_1, Come Back Later!

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.572399	0.208491	0.363908
cubberly-auditorium-2019-04-22_1	0.537999	0.159264	0.378735
discovery-walk-2019-02-28_0	0.620289	0.292631	0.327658
discovery-walk-2019-02-28_1	0.613622	0.241569	0.372053
food-trucks-2019-02-12_0	0.524448	0.127932	0.396516
gates-ai-lab-2019-04-17_0	0.678164	0.43903	0.239134
gates-basement-elevators-2019-01-17_0	0.487385	0.11382	0.373565
gates-foyer-2019-01-17_0	0.523641	0.208299	0.315342
gates-to-clark-2019-02-28_0	0.500013	0.167006	0.333007
hewlett-class-2019-01-23_0	0.44843	0.148515	0.299915
hewlett-class-2019-01-23_1	0.36809	0.066046	0.302044
huang-2-2019-01-25_1	0.608138	0.192735	0.415403
huang-intersection-2019-01-22_0	0.69776	0.375435	0.322325
indoor-coupa-cafe-2019-02-06_0	0.573818	0.191444	0.382374
lomita-serra-intersection-2019-01-30_0	0.641985	0.297623	0.344362
meyer-green-2019-03-16_1	0.706693	0.274197	0.432496
nvidia-aud-2019-01-25_0	0.62522	0.206343	0.418877
nvidia-aud-2019-04-18_1	0.571567	0.124179	0.447388
nvidia-aud-2019-04-18_2	0.553444	0.114655	0.438789
outdoor-coupa-cafe-2019-02-06_0	0.678975	0.254229	0.424745
quarry-road-2019-02-28_0	0.637471	0.157681	0.47979
serra-street-2019-01-30_0	0.742568	0.507667	0.234901
stlc-111-2019-04-19_1	0.401152	0.092255	0.308898
stlc-111-2019-04-19_2	0.422244	0.115204	0.30704
tressider-2019-03-16_2	0.6771	0.42857	0.248531
tressider-2019-04-26_0	0.564326	0.138205	0.426121
tressider-2019-04-26_1	0.496937	0.093629	0.403308
tressider-2019-04-26_3	0.553294	0.101107	0.452187

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.769048	0.460756	0.0529612
cubberly-auditorium-2019-04-22_1	0.842583	0.446877	0.0403997
discovery-walk-2019-02-28_0	0.950761	0.681955	0.114635
discovery-walk-2019-02-28_1	0.880158	0.560014	0.0527631
food-trucks-2019-02-12_0	0.874153	0.565468	0.05677
gates-ai-lab-2019-04-17_0	0.744601	0.582824	0.0986549
gates-basement-elevators-2019-01-17_0	0.835954	0.560879	0.0953749
gates-foyer-2019-01-17_0	0.87705	0.72018	0.145272
gates-to-clark-2019-02-28_0	0.943343	0.753396	0.225462
hewlett-class-2019-01-23_0	0.859942	0.659903	0.161235
hewlett-class-2019-01-23_1	0.960213	0.821572	0.223536
huang-2-2019-01-25_1	0.748432	0.502892	0.108067
huang-intersection-2019-01-22_0	0.901031	0.54318	0.0506058
indoor-coupa-cafe-2019-02-06_0	0.65621	0.410751	0.0383784
lomita-serra-intersection-2019-01-30_0	0.957535	0.741993	0.106959
meyer-green-2019-03-16_1	0.686628	0.390109	0.0948147
nvidia-aud-2019-01-25_0	0.695976	0.359326	0.0407643
nvidia-aud-2019-04-18_1	0.696539	0.544574	0.105778
nvidia-aud-2019-04-18_2	0.845181	0.519046	0.0752075
outdoor-coupa-cafe-2019-02-06_0	0.67548	0.288567	0.0160853
quarry-road-2019-02-28_0	0.854468	0.700704	0.186407
serra-street-2019-01-30_0	0.910634	0.540119	0.104599
stlc-111-2019-04-19_1	0.947729	0.835115	0.310385
stlc-111-2019-04-19_2	0.903999	0.776335	0.27719
tressider-2019-03-16_2	0.912261	0.578756	0.0292601
tressider-2019-04-26_0	0.719213	0.361453	0.0339652
tressider-2019-04-26_1	0.79894	0.485009	0.0288379
tressider-2019-04-26_3	0.717999	0.336109	0.0267766

Person-MinkUNet

0.582

76.351%

42.016%

2.649%

Dan Jia, and Bastian Leibe Person-MinkUNet: 3D Person Detection with LiDAR Point Cloud in CVPRW 2021

Details

Name	Person-MinkUNet
Submission Date	2022-02-16 18:25:51+00:00
Abstract	We take pointcloud from both upper and lower LiDARs, and voxelize them with 0.05x0.05x0.1 m voxels. The vowelized points are then passed into a Minkowski u-net backbone (implementation from https://github.com/mit-han-lab/spvnas/). We use a one-stage detector paradigm, where each active voxel directly generates a box, not using RPN or pooling. From all the boxes, a non-maximum-suppression is applied to obtain final detections. The network is trained for 40 epochs, with batch size 36 on a RTX TITAN. We use Adam optimizer, with a 1e-3 learning rate, with exponential decay to 1e-6 starting at the 15 epoch. For data augmentation, we used random scaling and random rotation along vertical axis.
Publication title	Person-MinkUNet: 3D Person Detection with LiDAR Point Cloud	Publication authors:	Dan Jia, and Bastian Leibe
Publication venue and year	CVPRW 2021	Publication URL:	https://arxiv.org/abs/2107.06780
Code Language	N/A	Hardware	1 TITAN RTX
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

Generating discovery-walk-2019-02-28_0, Come Back Later!

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.58185	0.199354	0.382496
cubberly-auditorium-2019-04-22_1	0.561116	0.181265	0.379852
discovery-walk-2019-02-28_0	0.646266	0.295966	0.3503
discovery-walk-2019-02-28_1	0.623724	0.24946	0.374264
food-trucks-2019-02-12_0	0.533644	0.122223	0.411422
gates-ai-lab-2019-04-17_0	0.610898	0.263936	0.346961
gates-basement-elevators-2019-01-17_0	0.49148	0.088885	0.402595
gates-foyer-2019-01-17_0	0.541926	0.185297	0.356629
gates-to-clark-2019-02-28_0	0.518009	0.157002	0.361007
hewlett-class-2019-01-23_0	0.467188	0.137612	0.329576
hewlett-class-2019-01-23_1	0.411509	0.071452	0.340057
huang-2-2019-01-25_1	0.571718	0.202009	0.369708
huang-intersection-2019-01-22_0	0.709133	0.409272	0.299861
indoor-coupa-cafe-2019-02-06_0	0.597458	0.176922	0.420536
lomita-serra-intersection-2019-01-30_0	0.668201	0.237287	0.430914
meyer-green-2019-03-16_1	0.713397	0.3275	0.385897
nvidia-aud-2019-01-25_0	0.639935	0.213144	0.42679
nvidia-aud-2019-04-18_1	0.550127	0.107273	0.442854
nvidia-aud-2019-04-18_2	0.549599	0.100533	0.449066
outdoor-coupa-cafe-2019-02-06_0	0.689522	0.268173	0.421349
quarry-road-2019-02-28_0	0.656847	0.17779	0.479058
serra-street-2019-01-30_0	0.750184	0.485534	0.264651
stlc-111-2019-04-19_1	0.430526	0.095804	0.334722
stlc-111-2019-04-19_2	0.459242	0.120328	0.338914
tressider-2019-03-16_2	0.687034	0.42166	0.265374
tressider-2019-04-26_0	0.577169	0.135841	0.441328
tressider-2019-04-26_1	0.504319	0.062331	0.441988
tressider-2019-04-26_3	0.549776	0.088065	0.461711

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.763509	0.420155	0.0264932
cubberly-auditorium-2019-04-22_1	0.830542	0.415399	0.0164996
discovery-walk-2019-02-28_0	0.947682	0.614585	0.0336624
discovery-walk-2019-02-28_1	0.877773	0.474343	0.0270392
food-trucks-2019-02-12_0	0.862094	0.510763	0.0424298
gates-ai-lab-2019-04-17_0	0.759885	0.544099	0.0384442
gates-basement-elevators-2019-01-17_0	0.866369	0.502471	0.048958
gates-foyer-2019-01-17_0	0.856094	0.614715	0.0561649
gates-to-clark-2019-02-28_0	0.936628	0.707403	0.100546
hewlett-class-2019-01-23_0	0.85338	0.580016	0.0831699
hewlett-class-2019-01-23_1	0.946324	0.75096	0.103055
huang-2-2019-01-25_1	0.799125	0.500119	0.0470769
huang-intersection-2019-01-22_0	0.903374	0.477458	0.0299851
indoor-coupa-cafe-2019-02-06_0	0.630921	0.34863	0.0148694
lomita-serra-intersection-2019-01-30_0	0.930039	0.632279	0.042619
meyer-green-2019-03-16_1	0.671941	0.327246	0.0439479
nvidia-aud-2019-01-25_0	0.666566	0.340798	0.024
nvidia-aud-2019-04-18_1	0.743816	0.543915	0.0985073
nvidia-aud-2019-04-18_2	0.846635	0.513551	0.0254766
outdoor-coupa-cafe-2019-02-06_0	0.669207	0.268729	0.00946484
quarry-road-2019-02-28_0	0.806995	0.592838	0.0692935
serra-street-2019-01-30_0	0.912844	0.460469	0.0383727
stlc-111-2019-04-19_1	0.939293	0.79606	0.130688
stlc-111-2019-04-19_2	0.886677	0.73163	0.116833
tressider-2019-03-16_2	0.846838	0.553939	0.011049
tressider-2019-04-26_0	0.708529	0.319923	0.0174577
tressider-2019-04-26_1	0.785846	0.450092	0.0199218
tressider-2019-04-26_3	0.715561	0.313172	0.0160946

PedNet

0.652

75.578%

20.238%

0.632%

Anonymous Submission

Details

Name	PedNet
Submission Date	2022-01-04 15:01:12+00:00
Abstract	Articulation modeling, feature extraction, and classification are the important components of pedestrian segmentation. Usually, these components are modeled independently from each other and then combined in a sequential way. However, this approach is prone to poor segmentation if any individual component is weakly designed. To cope with this problem, we proposed a spatio-temporal convolutional neural network named PedNet which exploits temporal information for spatial segmentation. The backbone of the PedNet consists of an encoder–decoder network for downsampling and upsampling the feature maps, respectively. The input to the network is a set of three frames and the output is a binary mask of the segmented regions in the middle frame. Irrespective of classical deep models where the convolution layers are followed by a fully connected layer for classification, PedNet is a Fully Convolutional Network (FCN). It is trained end-to-end and the segmentation is achieved without the need of any pre- or post-processing. The main characteristic of PedNet is its unique design where it performs segmentation on a frame-by-frame basis but it uses the temporal information from the previous and the future frame for segmenting the pedestrian in the current frame. Moreover, to combine the low-level features with the high-level semantic information learned by the deeper layers, we used long-skip connections from the encoder to decoder network and concatenate the output of low-level layers with the higher level layers. This approach helps to get segmentation map with sharp boundaries. To show the potential benefits of temporal information, we also visualized different layers of the network. The visualization showed that the network learned different information from the consecutive frames and then combined the information optimally to segment the middle frame. We evaluated our approach on eight challenging datasets where humans are involved in different activities with severe articulat
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	Python	Hardware	Titan V
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.651729	0.310081	0.341649
cubberly-auditorium-2019-04-22_1	0.647898	0.311826	0.336073
discovery-walk-2019-02-28_0	0.774846	0.526466	0.24838
discovery-walk-2019-02-28_1	0.686294	0.435685	0.25061
food-trucks-2019-02-12_0	0.621746	0.242475	0.379271
gates-ai-lab-2019-04-17_0	0.613672	0.150927	0.462745
gates-basement-elevators-2019-01-17_0	0.481293	0.141874	0.339419
gates-foyer-2019-01-17_0	0.550907	0.114943	0.435964
gates-to-clark-2019-02-28_0	0.587245	0.208528	0.378717
hewlett-class-2019-01-23_0	0.52439	0.09428	0.43011
hewlett-class-2019-01-23_1	0.513417	0.081804	0.431613
huang-2-2019-01-25_1	0.637198	0.255556	0.381642
huang-intersection-2019-01-22_0	0.846349	0.625576	0.220773
indoor-coupa-cafe-2019-02-06_0	0.648118	0.313151	0.334967
lomita-serra-intersection-2019-01-30_0	0.761267	0.467183	0.294085
meyer-green-2019-03-16_1	0.758213	0.396777	0.361436
nvidia-aud-2019-01-25_0	0.700867	0.370869	0.329998
nvidia-aud-2019-04-18_1	0.645252	0.188247	0.457004
nvidia-aud-2019-04-18_2	0.60466	0.227336	0.377323
outdoor-coupa-cafe-2019-02-06_0	0.740202	0.394106	0.346096
quarry-road-2019-02-28_0	0.714133	0.358319	0.355814
serra-street-2019-01-30_0	0.846993	0.728842	0.118151
stlc-111-2019-04-19_1	0.506711	0.106406	0.400305
stlc-111-2019-04-19_2	0.551521	0.157295	0.394226
tressider-2019-03-16_2	0.779429	0.640318	0.139111
tressider-2019-04-26_0	0.651737	0.327383	0.324353
tressider-2019-04-26_1	0.580998	0.246287	0.334711
tressider-2019-04-26_3	0.621334	0.259721	0.361613

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.755783	0.20238	0.00632289
cubberly-auditorium-2019-04-22_1	0.782588	0.194454	0.00337627
discovery-walk-2019-02-28_0	0.9345	0.156886	0.00388451
discovery-walk-2019-02-28_1	0.865509	0.442029	0.0412694
food-trucks-2019-02-12_0	0.857545	0.175538	0.00605504
gates-ai-lab-2019-04-17_0	0.762117	0.195287	0.00579208
gates-basement-elevators-2019-01-17_0	0.890555	0.449644	0.0735908
gates-foyer-2019-01-17_0	0.930615	0.227658	0.0101086
gates-to-clark-2019-02-28_0	0.910379	0.410841	0.0744382
hewlett-class-2019-01-23_0	0.81647	0.213154	0.0114972
hewlett-class-2019-01-23_1	0.96585	0.303006	0.01623
huang-2-2019-01-25_1	0.796694	0.309733	0.0140182
huang-intersection-2019-01-22_0	0.684234	0.0958863	0.00185602
indoor-coupa-cafe-2019-02-06_0	0.652381	0.17488	0.00292412
lomita-serra-intersection-2019-01-30_0	0.911123	0.426449	0.00613479
meyer-green-2019-03-16_1	0.648787	0.238372	0.0430074
nvidia-aud-2019-01-25_0	0.698107	0.143318	0.00330198
nvidia-aud-2019-04-18_1	0.686729	0.18284	0.00976622
nvidia-aud-2019-04-18_2	0.875319	0.315597	0.00427983
outdoor-coupa-cafe-2019-02-06_0	0.716391	0.104031	0.00106961
quarry-road-2019-02-28_0	0.778986	0.505488	0.117119
serra-street-2019-01-30_0	0.872858	0.331175	0.0466208
stlc-111-2019-04-19_1	0.935065	0.31062	0.0248626
stlc-111-2019-04-19_2	0.868904	0.328606	0.0308866
tressider-2019-03-16_2	0.86115	0.431463	0.0181987
tressider-2019-04-26_0	0.703941	0.162497	0.00291587
tressider-2019-04-26_1	0.762911	0.257466	0.00285661
tressider-2019-04-26_3	0.7078	0.135066	0.00272037

EPNet

0.653

63.589%

17.054%

0.604%

Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection in ECCV, 2020

Details

Name	EPNet
Submission Date	2021-11-24 17:26:08+00:00
Abstract	In this paper, we aim at addressing two critical issues in the 3D detection task, including the exploitation of multiple sensors (namely LiDAR point cloud and camera image), as well as the inconsistency between the localization and classification confidence.To this end, we propose a novel fusion module to enhance the point features with semantic image features in a point-wise manner without any image annotations. Besides, a consistency enforcing loss is employed to explicitly encourage the consistency of both the localization and classification confidence.
Publication title	EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection	Publication authors:	Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai
Publication venue and year	ECCV, 2020	Publication URL:	https://arxiv.org/pdf/2007.08856.pdf
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	https://github.com/happinesslz/EPNet

Visualisation

Best 3 By AP_0.3 ↓

lomita-serra-intersection-2019-01-30_0

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.652688	0.263658	0.38903
cubberly-auditorium-2019-04-22_1	0.64372	0.222753	0.420967
discovery-walk-2019-02-28_0	0.739471	0.421366	0.318105
discovery-walk-2019-02-28_1	0.693125	0.360835	0.33229
food-trucks-2019-02-12_0	0.599397	0.176981	0.422416
gates-ai-lab-2019-04-17_0	0.666251	0.297137	0.369114
gates-basement-elevators-2019-01-17_0	0.528054	0.096315	0.431739
gates-foyer-2019-01-17_0	0.629074	0.215912	0.413162
gates-to-clark-2019-02-28_0	0.544782	0.152743	0.392039
hewlett-class-2019-01-23_0	0.507062	0.111559	0.395503
hewlett-class-2019-01-23_1	0.487688	0.090365	0.397323
huang-2-2019-01-25_1	0.686527	0.287276	0.399251
huang-intersection-2019-01-22_0	0.760011	0.388467	0.371544
indoor-coupa-cafe-2019-02-06_0	0.674706	0.221484	0.453222
lomita-serra-intersection-2019-01-30_0	0.72	0.47146	0.248541
meyer-green-2019-03-16_1	0.751019	0.327183	0.423836
nvidia-aud-2019-01-25_0	0.700024	0.204972	0.495051
nvidia-aud-2019-04-18_1	0.691498	0.127031	0.564467
nvidia-aud-2019-04-18_2	0.618609	0.140855	0.477754
outdoor-coupa-cafe-2019-02-06_0	0.736419	0.311	0.425418
quarry-road-2019-02-28_0	0.706597	0.161788	0.544809
serra-street-2019-01-30_0	0.806544	0.616484	0.19006
stlc-111-2019-04-19_1	0.483998	0.086873	0.397125
stlc-111-2019-04-19_2	0.504334	0.117113	0.387222
tressider-2019-03-16_2	0.716924	0.416479	0.300445
tressider-2019-04-26_0	0.701601	0.407566	0.294035
tressider-2019-04-26_1	0.683749	0.415329	0.268421
tressider-2019-04-26_3	0.64139	0.27143	0.36996

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.635886	0.170543	0.0060391
cubberly-auditorium-2019-04-22_1	0.762286	0.160164	0.00288501
discovery-walk-2019-02-28_0	0.90068	0.231669	0.00566289
discovery-walk-2019-02-28_1	0.821669	0.279233	0.0103878
food-trucks-2019-02-12_0	0.825449	0.168776	0.00599767
gates-ai-lab-2019-04-17_0	0.758043	0.285097	0.0113628
gates-basement-elevators-2019-01-17_0	0.86038	0.27284	0.0146962
gates-foyer-2019-01-17_0	0.891658	0.252079	0.0182122
gates-to-clark-2019-02-28_0	0.91943	0.459405	0.0656653
hewlett-class-2019-01-23_0	0.783743	0.344538	0.0244957
hewlett-class-2019-01-23_1	0.935491	0.416189	0.0552076
huang-2-2019-01-25_1	0.669661	0.302079	0.027948
huang-intersection-2019-01-22_0	0.836635	0.152878	0.00449855
indoor-coupa-cafe-2019-02-06_0	0.557868	0.152107	0.00351947
lomita-serra-intersection-2019-01-30_0	0.944712	0.480768	0.021901
meyer-green-2019-03-16_1	0.633689	0.198133	0.0115316
nvidia-aud-2019-01-25_0	0.649178	0.108196	0.00225249
nvidia-aud-2019-04-18_1	0.653419	0.0663005	0.00210129
nvidia-aud-2019-04-18_2	0.808758	0.208564	0.00304954
outdoor-coupa-cafe-2019-02-06_0	0.619114	0.088884	0.000312417
quarry-road-2019-02-28_0	0.806433	0.556971	0.0771563
serra-street-2019-01-30_0	0.869388	0.262403	0.0331807
stlc-111-2019-04-19_1	0.904551	0.450065	0.0502385
stlc-111-2019-04-19_2	0.872955	0.462591	0.0865523
tressider-2019-03-16_2	0.815696	0.324522	0.00222187
tressider-2019-04-26_0	0.512586	0.143989	0.00356217
tressider-2019-04-26_1	0.492756	0.131917	0.00292005
tressider-2019-04-26_3	0.550831	0.136687	0.00392834

EPNet++

0.655

66.631%

18.661%

0.671%

Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection in ECCV, 2020

Details

Name	EPNet++
Submission Date	2021-11-24 23:32:56+00:00
Abstract	In this paper, we aim at addressing two critical issues in the 3D detection task, including the exploitation of multiple sensors (namely LiDAR point cloud and camera image), as well as the inconsistency be- tween the localization and classification confidence. To this end, we pro- pose a novel fusion module to enhance the point features with semantic image features in a point-wise manner. Besides, a consistency forcing loss is employed to explicitly encourage the consistency of both the localiza- tion and classification confidence. We design an end-to-end learnable framework named EPNet to integrate these two components. Extensive experiments on the KITTI and SUN-RGBD datasets demonstrate the superiority of EPNet over the state-of-the-art methods.
Publication title	EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection	Publication authors:	Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai
Publication venue and year	ECCV, 2020	Publication URL:	https://arxiv.org/pdf/2007.08856.pdf
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	https://github.com/happinesslz/EPNet

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.655091	0.229332	0.42576
cubberly-auditorium-2019-04-22_1	0.596234	0.130349	0.465885
discovery-walk-2019-02-28_0	0.724086	0.3377	0.386386
discovery-walk-2019-02-28_1	0.679163	0.178678	0.500485
food-trucks-2019-02-12_0	0.580694	0.089668	0.491026
gates-ai-lab-2019-04-17_0	0.708565	0.405431	0.303134
gates-basement-elevators-2019-01-17_0	0.539702	0.098068	0.441634
gates-foyer-2019-01-17_0	0.6923	0.36666	0.32564
gates-to-clark-2019-02-28_0	0.589575	0.20737	0.382205
hewlett-class-2019-01-23_0	0.51298	0.132423	0.380558
hewlett-class-2019-01-23_1	0.520936	0.109691	0.411245
huang-2-2019-01-25_1	0.669108	0.225061	0.444047
huang-intersection-2019-01-22_0	0.768529	0.21403	0.554498
indoor-coupa-cafe-2019-02-06_0	0.655238	0.170002	0.485237
lomita-serra-intersection-2019-01-30_0	0.727771	0.212884	0.514887
meyer-green-2019-03-16_1	0.743614	0.232667	0.510947
nvidia-aud-2019-01-25_0	0.688381	0.141529	0.546852
nvidia-aud-2019-04-18_1	0.756367	0.308964	0.447403
nvidia-aud-2019-04-18_2	0.650981	0.204858	0.446123
outdoor-coupa-cafe-2019-02-06_0	0.702263	0.160628	0.541635
quarry-road-2019-02-28_0	0.740532	0.254514	0.486019
serra-street-2019-01-30_0	0.786923	0.543899	0.243024
stlc-111-2019-04-19_1	0.509794	0.140927	0.368867
stlc-111-2019-04-19_2	0.546838	0.171308	0.37553
tressider-2019-03-16_2	0.681355	0.269306	0.41205
tressider-2019-04-26_0	0.667426	0.349613	0.317813
tressider-2019-04-26_1	0.635215	0.321482	0.313733
tressider-2019-04-26_3	0.612893	0.214245	0.398648

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.666313	0.186614	0.00670896
cubberly-auditorium-2019-04-22_1	0.806739	0.186916	0.00516292
discovery-walk-2019-02-28_0	0.922893	0.234548	0.00514681
discovery-walk-2019-02-28_1	0.840583	0.289015	0.0145141
food-trucks-2019-02-12_0	0.849237	0.205649	0.011993
gates-ai-lab-2019-04-17_0	0.760519	0.290246	0.00630754
gates-basement-elevators-2019-01-17_0	0.834842	0.291193	0.0155308
gates-foyer-2019-01-17_0	0.889946	0.250645	0.0128821
gates-to-clark-2019-02-28_0	0.925571	0.500114	0.097916
hewlett-class-2019-01-23_0	0.792839	0.357233	0.0326515
hewlett-class-2019-01-23_1	0.95852	0.358219	0.0327929
huang-2-2019-01-25_1	0.71466	0.377899	0.0404534
huang-intersection-2019-01-22_0	0.823571	0.109377	0.00167962
indoor-coupa-cafe-2019-02-06_0	0.611619	0.184545	0.00402507
lomita-serra-intersection-2019-01-30_0	0.905275	0.420877	0.0281802
meyer-green-2019-03-16_1	0.626977	0.208244	0.0131057
nvidia-aud-2019-01-25_0	0.652813	0.10986	0.00258273
nvidia-aud-2019-04-18_1	0.680664	0.0863731	0.000169866
nvidia-aud-2019-04-18_2	0.854472	0.219352	0.00257191
outdoor-coupa-cafe-2019-02-06_0	0.650742	0.0878304	0.000406002
quarry-road-2019-02-28_0	0.815685	0.561822	0.090299
serra-street-2019-01-30_0	0.891822	0.283562	0.0460585
stlc-111-2019-04-19_1	0.928024	0.451527	0.037146
stlc-111-2019-04-19_2	0.862916	0.450258	0.0823194
tressider-2019-03-16_2	0.861519	0.35177	0.00310532
tressider-2019-04-26_0	0.560883	0.168859	0.00384717
tressider-2019-04-26_1	0.568443	0.153727	0.00105314
tressider-2019-04-26_3	0.59913	0.156945	0.00380473

PiFeNet

0.66

74.284%

42.617%

4.886%

Duy-Tho Le, Hengcan Shi, Hamid Rezatofighi, Jianfei Cai Accurate and Real-time 3D Pedestrian Detection Using an Efficient Attentive Pillar Network in IEEE Robotics and Automation Letters

Details

Name	PiFeNet
Submission Date	2022-05-21 07:03:53+00:00
Abstract	Efficiently and accurately detecting people from 3D point cloud data is of great importance in many robotic and autonomous driving applications. This fundamental perception task is still very challenging due to (i) significant deformations of human body pose and gesture over time and (ii) point cloud sparsity and scarcity for pedestrian class objects. Recent efficient 3D object detection approaches rely on pillar features to detect objects from point cloud data. However, these pillar features do not carry sufficient expressive representations to deal with all the aforementioned challenges in detecting people. To address this shortcoming, we first introduce a stackable Pillar Aware Attention (PAA) module for enhanced pillar features extraction while suppressing noises in the point clouds. By integrating multi-point-channel-pooling, point-wise, channel-wise, and task-aware attention into a simple module, the representation capabilities are boosted while requiring little additional computing resources. We also present Mini-BiFPN, a small yet effective feature network that creates bidirectional information flow and multi-level cross-scale feature fusion to better integrate multi-resolution features. Our proposed framework, namely PiFeNet, has been evaluated on three popular large-scale datasets for 3D pedestrian Detection, i.e. KITTI, JRDB, and nuScenes achieving state-of-the-art (SOTA) performance on KITTI Bird-eye-view (BEV) and JRDB and very competitive performance on nuScenes. Our approach has inference speed of 26 frame-per-second (FPS), making it a real-time detector. The code for our PiFeNet is available at https://github.com/ldtho/PiFeNet
Publication title	Accurate and Real-time 3D Pedestrian Detection Using an Efficient Attentive Pillar Network	Publication authors:	Duy-Tho Le, Hengcan Shi, Hamid Rezatofighi, Jianfei Cai
Publication venue and year	IEEE Robotics and Automation Letters	Publication URL:	https://ieeexplore.ieee.org/abstract/document/10003992/
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

Generating hewlett-class-2019-01-23_1, Come Back Later!

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.660121	0.407483	0.252638
cubberly-auditorium-2019-04-22_1	0.669091	0.392806	0.276285
discovery-walk-2019-02-28_0	0.733474	0.547787	0.185687
discovery-walk-2019-02-28_1	0.668614	0.444642	0.223971
food-trucks-2019-02-12_0	0.646741	0.401236	0.245504
gates-ai-lab-2019-04-17_0	0.587211	0.13902	0.448191
gates-basement-elevators-2019-01-17_0	0.522806	0.204593	0.318213
gates-foyer-2019-01-17_0	0.541716	0.137704	0.404012
gates-to-clark-2019-02-28_0	0.51789	0.266519	0.251371
hewlett-class-2019-01-23_0	0.518461	0.217839	0.300622
hewlett-class-2019-01-23_1	0.434515	0.128296	0.306219
huang-2-2019-01-25_1	0.675215	0.455082	0.220134
huang-intersection-2019-01-22_0	0.776255	0.612886	0.163369
indoor-coupa-cafe-2019-02-06_0	0.728657	0.504652	0.224005
lomita-serra-intersection-2019-01-30_0	0.710489	0.50296	0.207529
meyer-green-2019-03-16_1	0.776625	0.593779	0.182845
nvidia-aud-2019-01-25_0	0.775492	0.559948	0.215545
nvidia-aud-2019-04-18_1	0.668235	0.295392	0.372843
nvidia-aud-2019-04-18_2	0.642939	0.375632	0.267308
outdoor-coupa-cafe-2019-02-06_0	0.784975	0.56004	0.224935
quarry-road-2019-02-28_0	0.690369	0.418367	0.272002
serra-street-2019-01-30_0	0.802285	0.678211	0.124074
stlc-111-2019-04-19_1	0.494338	0.181091	0.313247
stlc-111-2019-04-19_2	0.473464	0.151879	0.321585
tressider-2019-03-16_2	0.762351	0.616634	0.145718
tressider-2019-04-26_0	0.750664	0.543524	0.20714
tressider-2019-04-26_1	0.734219	0.539242	0.194977
tressider-2019-04-26_3	0.736168	0.53227	0.203898

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.74284	0.42617	0.0488638
cubberly-auditorium-2019-04-22_1	0.773657	0.414712	0.0374064
discovery-walk-2019-02-28_0	0.950476	0.636288	0.134042
discovery-walk-2019-02-28_1	0.881574	0.605271	0.12428
food-trucks-2019-02-12_0	0.857426	0.569187	0.0826523
gates-ai-lab-2019-04-17_0	0.737022	0.511669	0.0793242
gates-basement-elevators-2019-01-17_0	0.855162	0.586117	0.106286
gates-foyer-2019-01-17_0	0.882166	0.603302	0.104235
gates-to-clark-2019-02-28_0	0.930668	0.782771	0.371805
hewlett-class-2019-01-23_0	0.850887	0.582957	0.137617
hewlett-class-2019-01-23_1	0.952504	0.785119	0.274683
huang-2-2019-01-25_1	0.760584	0.498329	0.150423
huang-intersection-2019-01-22_0	0.85488	0.598126	0.0999539
indoor-coupa-cafe-2019-02-06_0	0.644913	0.308365	0.0176668
lomita-serra-intersection-2019-01-30_0	0.948186	0.793934	0.155099
meyer-green-2019-03-16_1	0.6708	0.412487	0.13655
nvidia-aud-2019-01-25_0	0.696359	0.334299	0.0285297
nvidia-aud-2019-04-18_1	0.612941	0.427075	0.0717368
nvidia-aud-2019-04-18_2	0.863558	0.54974	0.0629867
outdoor-coupa-cafe-2019-02-06_0	0.701849	0.289988	0.0158044
quarry-road-2019-02-28_0	0.869349	0.789577	0.354918
serra-street-2019-01-30_0	0.925837	0.603653	0.13033
stlc-111-2019-04-19_1	0.921748	0.79114	0.23961
stlc-111-2019-04-19_2	0.916671	0.745539	0.295884
tressider-2019-03-16_2	0.909394	0.648525	0.0254869
tressider-2019-04-26_0	0.686949	0.326921	0.0279434
tressider-2019-04-26_1	0.742617	0.423797	0.0168478
tressider-2019-04-26_3	0.70152	0.302553	0.0203075

EPNet_lidar

0.677

59.252%

16.845%

0.418%

Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai. EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection. in ECCV, 2020.

Details

Name	EPNet_lidar
Submission Date	2021-11-24 23:28:56+00:00
Abstract	In this paper, we aim at addressing two critical issues in the 3D detection task, including the exploitation of multiple sensors (namely LiDAR point cloud and camera image), as well as the inconsistency between the localization and classification confidence.To this end, we propose a novel fusion module to enhance the point features with semantic image features in a point-wise manner without any image annotations. Besides, a consistency enforcing loss is employed to explicitly encourage the consistency of both the localization and classification confidence.
Publication title	EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection.	Publication authors:	Tengteng Huang, Zhe Liu, Xiwu Chen, and Xiang Bai.
Publication venue and year	ECCV, 2020.	Publication URL:	https://arxiv.org/pdf/2007.08856.pdf
Code Language	N/A	Hardware	1 GPU(Titan V)
Code Website	N/A	Code URL	https://github.com/happinesslz/EPNet

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.677476	0.271097	0.406379
cubberly-auditorium-2019-04-22_1	0.638384	0.136795	0.501589
discovery-walk-2019-02-28_0	0.737789	0.374119	0.363669
discovery-walk-2019-02-28_1	0.676655	0.244268	0.432388
food-trucks-2019-02-12_0	0.611723	0.191912	0.419812
gates-ai-lab-2019-04-17_0	0.707912	0.348411	0.359501
gates-basement-elevators-2019-01-17_0	0.522206	0.088499	0.433707
gates-foyer-2019-01-17_0	0.733067	0.434757	0.29831
gates-to-clark-2019-02-28_0	0.621768	0.238281	0.383487
hewlett-class-2019-01-23_0	0.540084	0.142156	0.397929
hewlett-class-2019-01-23_1	0.5387	0.129278	0.409422
huang-2-2019-01-25_1	0.722799	0.211164	0.511636
huang-intersection-2019-01-22_0	0.757429	0.410704	0.346724
indoor-coupa-cafe-2019-02-06_0	0.659689	0.201939	0.45775
lomita-serra-intersection-2019-01-30_0	0.712779	0.25286	0.459919
meyer-green-2019-03-16_1	0.74898	0.238218	0.510762
nvidia-aud-2019-01-25_0	0.692088	0.103332	0.588757
nvidia-aud-2019-04-18_1	0.757571	0.333159	0.424412
nvidia-aud-2019-04-18_2	0.677344	0.178229	0.499115
outdoor-coupa-cafe-2019-02-06_0	0.718392	0.210263	0.508129
quarry-road-2019-02-28_0	0.735393	0.176457	0.558936
serra-street-2019-01-30_0	0.795524	0.549789	0.245735
stlc-111-2019-04-19_1	0.531297	0.123388	0.407909
stlc-111-2019-04-19_2	0.6081	0.254114	0.353986
tressider-2019-03-16_2	0.738298	0.42905	0.309247
tressider-2019-04-26_0	0.736419	0.498648	0.237771
tressider-2019-04-26_1	0.702029	0.464709	0.23732
tressider-2019-04-26_3	0.669424	0.35512	0.314304

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.592524	0.168454	0.0041769
cubberly-auditorium-2019-04-22_1	0.755045	0.202184	0.00289067
discovery-walk-2019-02-28_0	0.887992	0.279676	0.0100217
discovery-walk-2019-02-28_1	0.849797	0.297088	0.0119964
food-trucks-2019-02-12_0	0.757168	0.174392	0.0114952
gates-ai-lab-2019-04-17_0	0.687834	0.227303	0.00508918
gates-basement-elevators-2019-01-17_0	0.881297	0.254748	0.0126534
gates-foyer-2019-01-17_0	0.841168	0.213076	0.00858194
gates-to-clark-2019-02-28_0	0.89543	0.498757	0.116076
hewlett-class-2019-01-23_0	0.738318	0.297371	0.0223551
hewlett-class-2019-01-23_1	0.944201	0.321538	0.0320492
huang-2-2019-01-25_1	0.682878	0.316706	0.029377
huang-intersection-2019-01-22_0	0.832757	0.169231	0.00536847
indoor-coupa-cafe-2019-02-06_0	0.554418	0.155921	0.00297088
lomita-serra-intersection-2019-01-30_0	0.938251	0.467859	0.027417
meyer-green-2019-03-16_1	0.647129	0.213002	0.0273785
nvidia-aud-2019-01-25_0	0.616565	0.105771	0.0022796
nvidia-aud-2019-04-18_1	0.681321	0.125004	0.000484833
nvidia-aud-2019-04-18_2	0.770224	0.165874	0.00191084
outdoor-coupa-cafe-2019-02-06_0	0.636101	0.0896675	0.000395898
quarry-road-2019-02-28_0	0.781164	0.512857	0.0695911
serra-street-2019-01-30_0	0.887442	0.208698	0.0220352
stlc-111-2019-04-19_1	0.901797	0.419788	0.0270172
stlc-111-2019-04-19_2	0.811985	0.405019	0.0494776
tressider-2019-03-16_2	0.810315	0.322334	0.00370303
tressider-2019-04-26_0	0.466559	0.141071	0.0038516
tressider-2019-04-26_1	0.471544	0.133327	0.00115797
tressider-2019-04-26_3	0.502213	0.128743	0.0036832

F-ConvNet

0.709

39.781%

8.116%

0.186%

Anonymous Submission

Details

Name	F-ConvNet
Submission Date	2021-11-24 17:31:12+00:00
Abstract	Abstract— In this work, we propose a novel method termed Frustum ConvNet (F-ConvNet) for amodal 3D object detection from point clouds. Given 2D region proposals in an RGB image, our method first generates a sequence of frustums for each region proposal, and uses the obtained frustums to group local points. F-ConvNet aggregates point-wise features as frustumlevel feature vectors, and arrays these feature vectors as a feature map for use of its subsequent component of fully convolutional network (FCN), which spatially fuses frustum-level features and supports an end-to-end and continuous estimation of oriented boxes in the 3D space. We also propose component variants of F-ConvNet, including an FCN variant that extracts multi-resolution frustum features, and a refined use of FConvNet over a reduced 3D space. Careful ablation studies verify the efficacy of these component variants. F-ConvNet assumes no prior knowledge of the working 3D environment and is thus dataset-agnostic. We present experiments on both the indoor SUN-RGBD and outdoor KITTI datasets. F-ConvNet outperforms all existing methods on SUN-RGBD, and at the time of submission it outperforms all published works on the KITTI benchmark.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	https://github.com/Gorilla-Lab-SCUT/frustum-convnet

Visualisation

Best 3 By AP_0.3 ↓

stlc-111-2019-04-19_2

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.709021	0.223385	0.485636
cubberly-auditorium-2019-04-22_1	0.657617	0.196109	0.461507
discovery-walk-2019-02-28_0	0.816091	0.27813	0.537961
discovery-walk-2019-02-28_1	0.713425	0.247135	0.46629
food-trucks-2019-02-12_0	0.663473	0.210024	0.453449
gates-ai-lab-2019-04-17_0	0.73606	0.360121	0.375939
gates-basement-elevators-2019-01-17_0	0.638073	0.119797	0.518276
gates-foyer-2019-01-17_0	0.687727	0.234299	0.453428
gates-to-clark-2019-02-28_0	0.712706	0.263427	0.44928
hewlett-class-2019-01-23_0	0.584638	0.109589	0.47505
hewlett-class-2019-01-23_1	0.614374	0.162767	0.451608
huang-2-2019-01-25_1	0.726719	0.161839	0.56488
huang-intersection-2019-01-22_0	0.817821	0.192814	0.625007
indoor-coupa-cafe-2019-02-06_0	0.709532	0.198162	0.51137
lomita-serra-intersection-2019-01-30_0	0.76392	0.381811	0.382109
meyer-green-2019-03-16_1	0.7601	0.188045	0.572056
nvidia-aud-2019-01-25_0	0.744091	0.137321	0.60677
nvidia-aud-2019-04-18_1	0.747346	0.180391	0.566955
nvidia-aud-2019-04-18_2	0.670812	0.086446	0.584366
outdoor-coupa-cafe-2019-02-06_0	0.783237	0.154342	0.628895
quarry-road-2019-02-28_0	0.807086	0.345865	0.461221
serra-street-2019-01-30_0	0.826701	0.536718	0.289983
stlc-111-2019-04-19_1	0.569816	0.134842	0.434974
stlc-111-2019-04-19_2	0.574959	0.116303	0.458656
tressider-2019-03-16_2	0.773005	0.216918	0.556088
tressider-2019-04-26_0	0.715895	0.2993	0.416595
tressider-2019-04-26_1	0.664848	0.318102	0.346747
tressider-2019-04-26_3	0.663485	0.200783	0.462702

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.397807	0.0811645	0.00185823
cubberly-auditorium-2019-04-22_1	0.543881	0.138057	0.0032417
discovery-walk-2019-02-28_0	0.503794	0.0594254	0.000424549
discovery-walk-2019-02-28_1	0.500886	0.23028	0.017302
food-trucks-2019-02-12_0	0.596174	0.118753	0.00293568
gates-ai-lab-2019-04-17_0	0.476636	0.0962487	0.0022164
gates-basement-elevators-2019-01-17_0	0.465699	0.132759	0.00330898
gates-foyer-2019-01-17_0	0.609531	0.117663	0.00188111
gates-to-clark-2019-02-28_0	0.611199	0.310635	0.0358944
hewlett-class-2019-01-23_0	0.525273	0.0988875	0.00696725
hewlett-class-2019-01-23_1	0.696575	0.17395	0.00550327
huang-2-2019-01-25_1	0.433677	0.190993	0.0317283
huang-intersection-2019-01-22_0	0.492944	0.08343	0.00220969
indoor-coupa-cafe-2019-02-06_0	0.230959	0.0269271	0.000428553
lomita-serra-intersection-2019-01-30_0	0.694949	0.275534	0.0296841
meyer-green-2019-03-16_1	0.381203	0.120081	0.014456
nvidia-aud-2019-01-25_0	0.315438	0.0326578	0.000336912
nvidia-aud-2019-04-18_1	0.435469	0.0541178	0.000174887
nvidia-aud-2019-04-18_2	0.517472	0.117211	0.000498265
outdoor-coupa-cafe-2019-02-06_0	0.26504	0.0241977	0.000173205
quarry-road-2019-02-28_0	0.587019	0.338552	0.0415777
serra-street-2019-01-30_0	0.429725	0.190844	0.0213134
stlc-111-2019-04-19_1	0.70159	0.214904	0.0120354
stlc-111-2019-04-19_2	0.722244	0.244113	0.0207049
tressider-2019-03-16_2	0.555494	0.153521	0.00029625
tressider-2019-04-26_0	0.278263	0.0312957	0.000361613
tressider-2019-04-26_1	0.384707	0.108511	0.000667287
tressider-2019-04-26_3	0.363985	0.0368936	0.000407493

TANet++

0.732

63.922%

27.991%

1.842%

Cong Ma TANet++: Triple Attention Network with Filtered Pointcloud on 3D Detection in arXiv preprint arXiv:2106.1536 (2021)

Details

Name	TANet++
Submission Date	2021-11-24 23:22:17+00:00
Abstract	TANet is one of state-of-the-art 3D object detection method on KITTI and JRDB benchmark, the network contains a Triple Attention module and Coarse-to-Fine Regression module to improve the robustness and accuracy of 3D Detection. However, since the original input data (point clouds) contains a lot of noise during collecting the data, which will further affect the training of the model. For example, the object is far from the robot, the sensor is difficult to obtain enough pointcloud. If the objects only contains few point clouds, and the samples are fed into model with the normal samples together during training, the detector will be difficult to distinguish the individual with few pointcloud belong to object or background. In this paper, we propose TANet++ to improve the performance on 3D Detection, which adopt a novel training strategy on training the TANet. In order to reduce the negative impact by the weak samples, the training strategy previously filtered the training data, and then the TANet++ is trained by the rest of data. The experimental results shows that AP score of TANet++ is 8.98% higher than TANet on JRDB benchmark.
Publication title	TANet++: Triple Attention Network with Filtered Pointcloud on 3D Detection	Publication authors:	Cong Ma
Publication venue and year	arXiv preprint arXiv:2106.1536 (2021)	Publication URL:	https://arxiv.org/pdf/2106.15366.pdf
Code Language	N/A	Hardware	1 Titan Tesla K40c
Code Website	N/A	Code URL	https://github.com/happinesslz/TANet

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.731876	0.527285	0.203941
cubberly-auditorium-2019-04-22_1	0.756998	0.548801	0.200783
discovery-walk-2019-02-28_0	0.855754	0.738569	0.108793
discovery-walk-2019-02-28_1	0.748846	0.584015	0.164832
food-trucks-2019-02-12_0	0.747767	0.481905	0.265862
gates-ai-lab-2019-04-17_0	0.63683	0.262858	0.373971
gates-basement-elevators-2019-01-17_0	0.607156	0.364676	0.242479
gates-foyer-2019-01-17_0	0.580804	0.214009	0.366795
gates-to-clark-2019-02-28_0	0.600447	0.317091	0.283356
hewlett-class-2019-01-23_0	0.583562	0.290415	0.293147
hewlett-class-2019-01-23_1	0.527361	0.249273	0.278088
huang-2-2019-01-25_1	0.758723	0.514145	0.242821
huang-intersection-2019-01-22_0	0.849857	0.729019	0.120837
indoor-coupa-cafe-2019-02-06_0	0.835734	0.715549	0.120185
lomita-serra-intersection-2019-01-30_0	0.804689	0.619793	0.184895
meyer-green-2019-03-16_1	0.841179	0.723557	0.117622
nvidia-aud-2019-01-25_0	0.837694	0.691905	0.145789
nvidia-aud-2019-04-18_1	0.710177	0.451997	0.25818
nvidia-aud-2019-04-18_2	0.735791	0.512259	0.223533
outdoor-coupa-cafe-2019-02-06_0	0.865736	0.728275	0.137461
quarry-road-2019-02-28_0	0.765271	0.576851	0.18842
serra-street-2019-01-30_0	0.881708	0.819017	0.062691
stlc-111-2019-04-19_1	0.577114	0.34138	0.235734
stlc-111-2019-04-19_2	0.574839	0.322227	0.252612
tressider-2019-03-16_2	0.80673	0.726854	0.079875
tressider-2019-04-26_0	0.814435	0.664295	0.15014
tressider-2019-04-26_1	0.707811	0.500168	0.207643
tressider-2019-04-26_3	0.747644	0.547781	0.199863

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.63922	0.279913	0.0184158
cubberly-auditorium-2019-04-22_1	0.709646	0.27542	0.0133194
discovery-walk-2019-02-28_0	0.658285	0.333364	0.0216991
discovery-walk-2019-02-28_1	0.691434	0.476388	0.063575
food-trucks-2019-02-12_0	0.745069	0.307807	0.00724537
gates-ai-lab-2019-04-17_0	0.661827	0.351595	0.0198363
gates-basement-elevators-2019-01-17_0	0.805537	0.460202	0.0577393
gates-foyer-2019-01-17_0	0.824157	0.45628	0.0325136
gates-to-clark-2019-02-28_0	0.75401	0.65871	0.220286
hewlett-class-2019-01-23_0	0.819215	0.441507	0.0415009
hewlett-class-2019-01-23_1	0.915634	0.701258	0.107298
huang-2-2019-01-25_1	0.65003	0.340237	0.0790799
huang-intersection-2019-01-22_0	0.626508	0.322855	0.0436361
indoor-coupa-cafe-2019-02-06_0	0.537171	0.168547	0.00332209
lomita-serra-intersection-2019-01-30_0	0.700028	0.405522	0.100227
meyer-green-2019-03-16_1	0.518292	0.276227	0.0692321
nvidia-aud-2019-01-25_0	0.50542	0.202102	0.0123053
nvidia-aud-2019-04-18_1	0.576044	0.319613	0.0360873
nvidia-aud-2019-04-18_2	0.674196	0.275347	0.0201017
outdoor-coupa-cafe-2019-02-06_0	0.537497	0.167933	0.00694234
quarry-road-2019-02-28_0	0.701916	0.605786	0.203305
serra-street-2019-01-30_0	0.623736	0.390883	0.053611
stlc-111-2019-04-19_1	0.893132	0.679667	0.113668
stlc-111-2019-04-19_2	0.834973	0.588418	0.148859
tressider-2019-03-16_2	0.698366	0.526818	0.122157
tressider-2019-04-26_0	0.614667	0.184083	0.00463452
tressider-2019-04-26_1	0.669851	0.297284	0.00841207
tressider-2019-04-26_3	0.614804	0.197299	0.0075886

F-PointNet

0.764

38.205%

6.378%

0.081%

C. Qi, W. Liu, C. Wu, H. Su and L. Guibas. Frustum PointNets for 3D Object Detection from RGB-D Data. in CVPR, 2018

Details

Name	F-PointNet
Submission Date	2021-11-24 23:26:56+00:00
Abstract	In this work, we study 3D object detection from RGB-D data in both indoor and outdoor scenes. While previous methods focus on images or 3D voxels, often obscuring natural 3D patterns and invariances of 3D data, we directly operate on raw point clouds by popping up RGB-D scans. However, a key challenge of this approach is how to efficiently localize objects in point clouds of large-scale scenes (region proposal). Instead of solely relying on 3D proposals, our method leverages both mature 2D object detectors and advanced 3D deep learning for object localization, achieving efficiency as well as high recall for even small objects. Benefited from learning directly in raw point clouds, our method is also able to precisely estimate 3D bounding boxes even under strong occlusion or with very sparse points. Evaluated on KITTI and SUN RGB-D 3D detection benchmarks, our method outperforms the state of the art by remarkable margins while having real-time capability.
Publication title	Frustum PointNets for 3D Object Detection from RGB-D Data.	Publication authors:	C. Qi, W. Liu, C. Wu, H. Su and L. Guibas.
Publication venue and year	CVPR, 2018	Publication URL:	https://arxiv.org/abs/1711.08488
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	https://github.com/charlesq34/frustum-pointnets

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.763534	0.445537	0.317997
cubberly-auditorium-2019-04-22_1	0.758388	0.462784	0.295604
discovery-walk-2019-02-28_0	0.834853	0.602754	0.2321
discovery-walk-2019-02-28_1	0.802284	0.518725	0.283559
food-trucks-2019-02-12_0	0.778525	0.555553	0.222972
gates-ai-lab-2019-04-17_0	0.675729	0.120903	0.554826
gates-basement-elevators-2019-01-17_0	0.7062	0.336408	0.369792
gates-foyer-2019-01-17_0	0.650699	0.160465	0.490234
gates-to-clark-2019-02-28_0	0.708545	0.281583	0.426962
hewlett-class-2019-01-23_0	0.594135	0.185739	0.408396
hewlett-class-2019-01-23_1	0.593223	0.121179	0.472043
huang-2-2019-01-25_1	0.769271	0.360161	0.40911
huang-intersection-2019-01-22_0	0.846841	0.629965	0.216876
indoor-coupa-cafe-2019-02-06_0	0.796187	0.550074	0.246113
lomita-serra-intersection-2019-01-30_0	0.812136	0.60176	0.210376
meyer-green-2019-03-16_1	0.888373	0.681853	0.20652
nvidia-aud-2019-01-25_0	0.82562	0.520938	0.304682
nvidia-aud-2019-04-18_1	0.702305	0.293777	0.408528
nvidia-aud-2019-04-18_2	0.709053	0.20716	0.501893
outdoor-coupa-cafe-2019-02-06_0	0.904793	0.72148	0.183313
quarry-road-2019-02-28_0	0.820217	0.387128	0.433088
serra-street-2019-01-30_0	0.883696	0.730473	0.153223
stlc-111-2019-04-19_1	0.638245	0.245642	0.392603
stlc-111-2019-04-19_2	0.660854	0.200754	0.460101
tressider-2019-03-16_2	0.819497	0.660168	0.15933
tressider-2019-04-26_0	0.848704	0.67622	0.172483
tressider-2019-04-26_1	0.777661	0.583542	0.19412
tressider-2019-04-26_3	0.809378	0.632309	0.177069

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.382051	0.0637768	0.000812578
cubberly-auditorium-2019-04-22_1	0.450084	0.0842632	0.000346808
discovery-walk-2019-02-28_0	0.54314	0.0879037	0.000536232
discovery-walk-2019-02-28_1	0.60324	0.0192989	0.000239707
food-trucks-2019-02-12_0	0.442772	0.111494	0.00100731
gates-ai-lab-2019-04-17_0	0.553188	0.104608	0.000574961
gates-basement-elevators-2019-01-17_0	0.477009	0.0829434	0.00284761
gates-foyer-2019-01-17_0	0.6986	0.180808	0.00192722
gates-to-clark-2019-02-28_0	0.666941	0.0938655	0.000607549
hewlett-class-2019-01-23_0	0.577646	0.14102	0.0121803
hewlett-class-2019-01-23_1	0.842756	0.106298	0.00241472
huang-2-2019-01-25_1	0.533702	0.0265473	0.000573034
huang-intersection-2019-01-22_0	0.531303	0.0788601	0.00233683
indoor-coupa-cafe-2019-02-06_0	0.303315	0.0395341	0.000556656
lomita-serra-intersection-2019-01-30_0	0.710519	0.203222	0.000628109
meyer-green-2019-03-16_1	0.314786	0.00680246	0.00018929
nvidia-aud-2019-01-25_0	0.345765	0.0293868	0.000744004
nvidia-aud-2019-04-18_1	0.529262	0.124015	0.0006852
nvidia-aud-2019-04-18_2	0.570361	0.129628	0.000492149
outdoor-coupa-cafe-2019-02-06_0	0.20002	0.0185828	0.000530909
quarry-road-2019-02-28_0	0.627778	0.0201335	0.000213068
serra-street-2019-01-30_0	0.517037	0.0356082	0.000385361
stlc-111-2019-04-19_1	0.665362	0.11734	0.00072367
stlc-111-2019-04-19_2	0.700961	0.074301	0.000597787
tressider-2019-03-16_2	0.554056	0.150497	0.000584302
tressider-2019-04-26_0	0.224149	0.0268836	0.000549875
tressider-2019-04-26_1	0.386433	0.0817232	0.00126349
tressider-2019-04-26_3	0.257069	0.0572863	0.000936715

TANet_on_JRDB

0.788

53.867%

4.175%

0.01%

Anonymous Submission

Details

Name	TANet_on_JRDB
Submission Date	2021-11-24 23:24:42+00:00
Abstract	conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.conduct 3d detection on JRDB with TANet.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

stlc-111-2019-04-19_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.787506	0.496602	0.290594
cubberly-auditorium-2019-04-22_1	0.803172	0.56234	0.232491
discovery-walk-2019-02-28_0	0.843527	0.64997	0.193557
discovery-walk-2019-02-28_1	0.796468	0.488537	0.307931
food-trucks-2019-02-12_0	0.77519	0.475269	0.29992
gates-ai-lab-2019-04-17_0	0.66136	0.164228	0.497132
gates-basement-elevators-2019-01-17_0	0.717476	0.360752	0.356725
gates-foyer-2019-01-17_0	0.686157	0.224515	0.461642
gates-to-clark-2019-02-28_0	0.739371	0.279743	0.459627
hewlett-class-2019-01-23_0	0.662154	0.267894	0.39426
hewlett-class-2019-01-23_1	0.691465	0.193443	0.498022
huang-2-2019-01-25_1	0.820506	0.458291	0.362215
huang-intersection-2019-01-22_0	0.880822	0.719885	0.160937
indoor-coupa-cafe-2019-02-06_0	0.825255	0.635622	0.189633
lomita-serra-intersection-2019-01-30_0	0.81595	0.499418	0.316533
meyer-green-2019-03-16_1	0.890879	0.663048	0.22783
nvidia-aud-2019-01-25_0	0.856262	0.637611	0.218651
nvidia-aud-2019-04-18_1	0.809296	0.571497	0.237799
nvidia-aud-2019-04-18_2	0.816266	0.540631	0.275635
outdoor-coupa-cafe-2019-02-06_0	0.880658	0.692815	0.187843
quarry-road-2019-02-28_0	0.815505	0.441857	0.373648
serra-street-2019-01-30_0	0.886666	0.749718	0.136947
stlc-111-2019-04-19_1	0.698519	0.373785	0.324733
stlc-111-2019-04-19_2	0.731894	0.386321	0.345573
tressider-2019-03-16_2	0.827088	0.681106	0.145982
tressider-2019-04-26_0	0.827145	0.655834	0.171311
tressider-2019-04-26_1	0.734091	0.499879	0.234212
tressider-2019-04-26_3	0.769507	0.534248	0.235259

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.538667	0.0417452	9.87144e-05
cubberly-auditorium-2019-04-22_1	0.699385	0.0462995	9.4882e-05
discovery-walk-2019-02-28_0	0.748175	0.0711298	0.00110513
discovery-walk-2019-02-28_1	0.727034	0.00724883	7.44455e-05
food-trucks-2019-02-12_0	0.614441	0.0498202	0.000120684
gates-ai-lab-2019-04-17_0	0.594104	0.0935583	0.00126395
gates-basement-elevators-2019-01-17_0	0.715344	0.00870945	7.25435e-05
gates-foyer-2019-01-17_0	0.713014	0.0832039	0.00123349
gates-to-clark-2019-02-28_0	0.79339	0.0165427	5.11533e-05
hewlett-class-2019-01-23_0	0.703504	0.0676289	0.00240993
hewlett-class-2019-01-23_1	0.780363	0.0264337	0.000273193
huang-2-2019-01-25_1	0.549004	0.00594769	2.02734e-05
huang-intersection-2019-01-22_0	0.577405	0.0263339	8.18587e-05
indoor-coupa-cafe-2019-02-06_0	0.385055	0.0270464	6.99458e-05
lomita-serra-intersection-2019-01-30_0	0.794898	0.0874336	0.00137793
meyer-green-2019-03-16_1	0.417536	0.00241434	1.5184e-05
nvidia-aud-2019-01-25_0	0.410571	0.0259448	9.44113e-05
nvidia-aud-2019-04-18_1	0.509932	0.0399049	0.0011087
nvidia-aud-2019-04-18_2	0.576085	0.0878814	0.00106666
outdoor-coupa-cafe-2019-02-06_0	0.390996	0.0204416	4.31537e-05
quarry-road-2019-02-28_0	0.764143	0.00871811	0.000101613
serra-street-2019-01-30_0	0.704275	0.0187657	0.000262213
stlc-111-2019-04-19_1	0.821682	0.0531515	0.00157808
stlc-111-2019-04-19_2	0.698376	0.0390169	0.000323741
tressider-2019-03-16_2	0.761704	0.123083	6.72536e-05
tressider-2019-04-26_0	0.475044	0.0397603	0.000259724
tressider-2019-04-26_1	0.565997	0.0729151	0.000297899
tressider-2019-04-26_3	0.483978	0.0349402	0.000228644

Team_MJM

0.828

69.204%

16.7%

0.359%

Anonymous Submission

Details

Name	Team_MJM
Submission Date	2021-11-24 17:23:30+00:00
Abstract	We focus on exploring the robustness of the3D object detection in point clouds, which has been rarelydiscussed in existing approaches. We observe two crucialphenomena: 1) the detection accuracy of the hard objects,e.g., Pedestrians, is unsatisfactory, 2) when adding additionalnoise points, the performance of existing approaches de-creases rapidly. To alleviate these problems, a novel TANet isintroduced in this paper, which mainly contains a Triple At-tention (TA) module, and a Coarse-to-Fine Regression (CFR)module. By considering the channel-wise, point-wise andvoxel-wise attention jointly, the TA module enhances the cru-cial information of the target while suppresses the unsta-ble cloud points. Besides, the novel stacked TA further ex-ploits the multi-level feature attention. In addition, the CFRmodule boosts the accuracy of localization without excessivecomputation cost. Experimental results on the validation setof KITTI dataset demonstrate that, in the challenging noisycases, i.e., adding additional random noisy points around eachobject, the presented approach goes far beyond state-of-the-art approaches. Furthermore, for the 3D object detection taskof the KITTI benchmark, our approach ranks the first place onPedestrian class, by using the point clouds as the only input.The running speed is around 29 frames per second.
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

stlc-111-2019-04-19_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.828337	0.673465	0.14042
cubberly-auditorium-2019-04-22_1	0.815692	0.597671	0.154072
discovery-walk-2019-02-28_0	0.889032	0.72185	0.130818
discovery-walk-2019-02-28_1	0.806163	0.680466	0.115558
food-trucks-2019-02-12_0	0.866266	0.735517	0.130749
gates-ai-lab-2019-04-17_0	0.784998	0.558412	0.226586
gates-basement-elevators-2019-01-17_0	0.675128	0.441668	0.233461
gates-foyer-2019-01-17_0	0.736431	0.414563	0.294809
gates-to-clark-2019-02-28_0	0.722137	0.463216	0.183628
hewlett-class-2019-01-23_0	0.755063	0.531992	0.223071
hewlett-class-2019-01-23_1	0.667208	0.442361	0.224847
huang-2-2019-01-25_1	0.843743	0.623979	0.119589
huang-intersection-2019-01-22_0	0.879623	0.763186	0.116436
indoor-coupa-cafe-2019-02-06_0	0.894807	0.815673	0.07194
lomita-serra-intersection-2019-01-30_0	0.879855	0.740866	0.111801
meyer-green-2019-03-16_1	0.883354	0.795769	0.079633
nvidia-aud-2019-01-25_0	0.88805	0.76518	0.120425
nvidia-aud-2019-04-18_1	0.868661	0.714273	0.134508
nvidia-aud-2019-04-18_2	0.843813	0.691554	0.15226
outdoor-coupa-cafe-2019-02-06_0	0.892684	0.777902	0.114782
quarry-road-2019-02-28_0	0.826232	0.700697	0.125536
serra-street-2019-01-30_0	0.832849	0.730623	0.102226
stlc-111-2019-04-19_1	0.759417	0.571013	0.188404
stlc-111-2019-04-19_2	0.770654	0.600253	0.164085
tressider-2019-03-16_2	0.881007	0.818171	0.056605
tressider-2019-04-26_0	0.91484	0.852406	0.062434
tressider-2019-04-26_1	0.877222	0.785264	0.091958
tressider-2019-04-26_3	0.910161	0.849035	0.061126

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.692043	0.167001	0.00358993
cubberly-auditorium-2019-04-22_1	0.785353	0.270422	0.00608185
discovery-walk-2019-02-28_0	0.862199	0.0972158	0.00201967
discovery-walk-2019-02-28_1	0.707604	0.326789	0.0218845
food-trucks-2019-02-12_0	0.821912	0.134769	0.0020016
gates-ai-lab-2019-04-17_0	0.691086	0.17046	0.00212645
gates-basement-elevators-2019-01-17_0	0.833876	0.343809	0.0183942
gates-foyer-2019-01-17_0	0.811844	0.187271	0.00411691
gates-to-clark-2019-02-28_0	0.722419	0.45437	0.0955063
hewlett-class-2019-01-23_0	0.812821	0.20874	0.0116014
hewlett-class-2019-01-23_1	0.914321	0.546025	0.0447723
huang-2-2019-01-25_1	0.698431	0.323245	0.0505198
huang-intersection-2019-01-22_0	0.71033	0.102929	0.00318967
indoor-coupa-cafe-2019-02-06_0	0.598931	0.117248	0.00126354
lomita-serra-intersection-2019-01-30_0	0.749483	0.274778	0.0382347
meyer-green-2019-03-16_1	0.570703	0.232581	0.0353728
nvidia-aud-2019-01-25_0	0.607331	0.0859134	0.00270398
nvidia-aud-2019-04-18_1	0.644652	0.164687	0.00300305
nvidia-aud-2019-04-18_2	0.745971	0.202115	0.0016404
outdoor-coupa-cafe-2019-02-06_0	0.605767	0.0957565	0.000346497
quarry-road-2019-02-28_0	0.697537	0.464368	0.16248
serra-street-2019-01-30_0	0.898434	0.471408	0.0816912
stlc-111-2019-04-19_1	0.930737	0.252996	0.0221439
stlc-111-2019-04-19_2	0.875322	0.393346	0.0584839
tressider-2019-03-16_2	0.679754	0.323123	0.00741327
tressider-2019-04-26_0	0.627377	0.125196	0.0021639
tressider-2019-04-26_1	0.703741	0.194489	0.000623099
tressider-2019-04-26_3	0.652244	0.118872	0.00216477

Team_minjunmin

0.918

57.262%

8.963%

0.157%

Anonymous Submission

Details

Name	Team_minjunmin
Submission Date	2021-11-24 17:27:48+00:00
Abstract	we focus on exploring the robustness of the3D object detection in point clouds, which has been rarelydiscussed in existing approaches. We observe two crucialphenomena: 1) the detection accuracy of the hard objects,e.g., Pedestrians, is unsatisfactory, 2) when adding additionalnoise points, the performance of existing approaches de-creases rapidly. To alleviate these problems, a novel TANet isintroduced in this paper, which mainly contains a Triple At-tention (TA) module, and a Coarse-to-Fine Regression (CFR)module. By considering the channel-wise, point-wise andvoxel-wise attention jointly, the TA module enhances the cru-cial information of the target while suppresses the unsta-ble cloud points. Besides, the novel stacked TA further ex-ploits the multi-level feature attention. In addition, the CFRmodule boosts the accuracy of localization without excessivecomputation cost. Experimental results on the validation setof KITTI dataset demonstrate that, in the challenging noisycases, i.e., adding additional random noisy points around eachobject, the presented approach goes far beyond state-of-the-art approaches. Furthermore, for the 3D object detection taskof the KITTI benchmark, our approach ranks the first place onPedestrian class, by using the point clouds as the only input.The running speed is around 29 frames per secon
Publication title	N/A	Publication authors:	N/A
Publication venue and year	N/A	Publication URL:	N/A
Code Language	N/A	Hardware	1 GPU (GTX 1080Ti)
Code Website	N/A	Code URL	N/A

Visualisation

Best 3 By AP_0.3 ↓

hewlett-class-2019-01-23_1

Worst 3 By AP_0.3 ↓

Per-sequence Results

Sequence	OSPA	OSPA_LOC	OSPA_CARD
overall	0.917936	0.686093	0.102813
cubberly-auditorium-2019-04-22_1	0.945686	0.607204	0.05674
discovery-walk-2019-02-28_0	0.985183	0.417097	0.017036
discovery-walk-2019-02-28_1	0.956873	0.649343	0.029964
food-trucks-2019-02-12_0	0.94774	0.880878	0.062498
gates-ai-lab-2019-04-17_0	0.899187	0.737216	0.157061
gates-basement-elevators-2019-01-17_0	0.869041	0.722914	0.132021
gates-foyer-2019-01-17_0	0.781949	0.439912	0.335422
gates-to-clark-2019-02-28_0	0.90073	0.468201	0.096059
hewlett-class-2019-01-23_0	0.906065	0.803446	0.084919
hewlett-class-2019-01-23_1	0.777553	0.576558	0.187018
huang-2-2019-01-25_1	0.892518	0.594319	0.280625
huang-intersection-2019-01-22_0	0.975602	0.650046	0.033211
indoor-coupa-cafe-2019-02-06_0	0.9198	0.822751	0.09585
lomita-serra-intersection-2019-01-30_0	0.972713	0.585658	0.030215
meyer-green-2019-03-16_1	0.957759	0.748966	0.047759
nvidia-aud-2019-01-25_0	0.919374	0.791787	0.123512
nvidia-aud-2019-04-18_1	0.903539	0.753308	0.122398
nvidia-aud-2019-04-18_2	0.852337	0.64447	0.207867
outdoor-coupa-cafe-2019-02-06_0	0.944432	0.757481	0.117296
quarry-road-2019-02-28_0	0.975864	0.479192	0.028931
serra-street-2019-01-30_0	0.927249	0.885215	0.034171
stlc-111-2019-04-19_1	0.901252	0.730665	0.102314
stlc-111-2019-04-19_2	0.908587	0.647718	0.164026
tressider-2019-03-16_2	0.976197	0.570433	0.002337
tressider-2019-04-26_0	0.925432	0.844038	0.081394
tressider-2019-04-26_1	0.939942	0.874355	0.065587
tressider-2019-04-26_3	0.921675	0.84134	0.079732

Sequence	AP_0.3	AP_0.5	AP_0.7
overall	0.57262	0.0896258	0.00157329
cubberly-auditorium-2019-04-22_1	0.737335	0.132587	0.0013431
discovery-walk-2019-02-28_0	0.69527	0.0650053	0.00108548
discovery-walk-2019-02-28_1	0.596967	0.184724	0.00454086
food-trucks-2019-02-12_0	0.676792	0.119579	0.000362015
gates-ai-lab-2019-04-17_0	0.527557	0.122333	0.00268597
gates-basement-elevators-2019-01-17_0	0.680004	0.0913794	0.00198108
gates-foyer-2019-01-17_0	0.650101	0.121875	0.00216985
gates-to-clark-2019-02-28_0	0.5088	0.25219	0.0273195
hewlett-class-2019-01-23_0	0.783555	0.18991	0.00632654
hewlett-class-2019-01-23_1	0.90934	0.451208	0.0328673
huang-2-2019-01-25_1	0.436843	0.0920824	0.0229082
huang-intersection-2019-01-22_0	0.512194	0.0542556	0.00104148
indoor-coupa-cafe-2019-02-06_0	0.560983	0.0705842	0.000300282
lomita-serra-intersection-2019-01-30_0	0.530941	0.104922	0.00479123
meyer-green-2019-03-16_1	0.439266	0.0427164	0.00704818
nvidia-aud-2019-01-25_0	0.525444	0.042457	0.000292732
nvidia-aud-2019-04-18_1	0.530684	0.0877624	0.00561484
nvidia-aud-2019-04-18_2	0.819395	0.11398	0.00180643
outdoor-coupa-cafe-2019-02-06_0	0.470216	0.0282515	0.000209964
quarry-road-2019-02-28_0	0.532776	0.20844	0.0294946
serra-street-2019-01-30_0	0.815933	0.333324	0.0470386
stlc-111-2019-04-19_1	0.818315	0.179708	0.00741579
stlc-111-2019-04-19_2	0.732863	0.150019	0.0140775
tressider-2019-03-16_2	0.614294	0.112032	0.000134645
tressider-2019-04-26_0	0.589827	0.0808507	0.00148566
tressider-2019-04-26_1	0.603319	0.129282	0.000499261
tressider-2019-04-26_3	0.531408	0.0566246	0.000476327

Symbol	Description
Individual Image	Method uses individual images from each camera
Stitched Image	Method uses stitched images combined from the individual cameras
Pointcloud	Method uses 3D pointcloud data
Online Tracking	Method does frame-by-frame processing with no lookahead
Offline Tracking	Method does not do in-order frame processing
Public Detections	Method uses publicly available detections
Private Detections	Method uses its own private detections

Measure	Better	Perfect	Description
OSPA
OSPA₂	lower	0.0	OSPA is a set-based metric which can directly capture a distance, between two sets of trajectories without a thresholding parameter[2].
OSPA Localization	lower	0.0	Representing different tracking error such as the displacement and size errors, track ID switches, track fragmentation or even track late initiation/early termination[2].
OSPA Cardinality	lower	0.0	Representing cardinality mismatch between two sets, penalizing missed or false tracks without an explicit definition for them [2].
AP
AP_0.3	higher	100%	Average Precision with intersection-over-union of bounding boxes larger than 30% [3].
AP_0.5	higher	100%	Average Precision with intersection-over-union of bounding boxes larger than 50% [3].
AP_0.7	higher	100%	Average Precision with intersection-over-union of bounding boxes larger than 70% [3].

Leaderboard

Instructions

2D Detection Submissions

3D Detection Submissions

Additional Information Used

Evaluation Measures[1]

Reference