JRDB Leaderboard

Instructions

This page display the submited results for Pose Leaderboards. For each submission, we display several main metrics in main table. For detailed information, more metrics, per-sequence results and visualisation (coming soon), please click submission name. For all tables, you can click headers to sort the results. Note you can download the submission zip file as well. Legends, metrics descriptions and reference are displayed after leaderboards table. For more information on submission preparation, click here.

2D Pose Estimation Submissions

Name

OSPA_IoU↓

AP_0.5 ↑

BUPT-COST_ECCV2022

0.303

64.047%

Jiajun Fu, Yonghao Dang, Ruoqi Yin, Shaojie Zhang, Feng Zhou, Wending Zhao, Jianqin Yin An Improved Baseline Framework for Pose Estimation Challenge at ECCV 2022 Visual Perception for Navi in Arxiv

Details

Name	BUPT-COST_ECCV2022
Submission Date	2022-10-10 14:35:51+00:00
Abstract	We directly predict human poses from the stitched images. We adopt HRNet-w48-384x288 from mmpose. For the train-valid data split, we use data from 'Bytes-coffee, ’Huang‘, ’Huang-hall‘, 'Huang-lane', and 'Jordan-hall' as the validation set, while the data from the rest locations are used as the training set. For human pose joints, we filter some instances, which lie on both sides of images. Besides, we obtain the bounding box based on the range of individual human pose joints during training. The team members are listed below. Advisor: Jianqin Yin Members: Jiajun Fu, Shaojie Zhang, Feng Zhou, Wending Zhao, Yonghao Dang, Ruoqi Yin
Publication title	An Improved Baseline Framework for Pose Estimation Challenge at ECCV 2022 Visual Perception for Navi	Publication authors:	Jiajun Fu, Yonghao Dang, Ruoqi Yin, Shaojie Zhang, Feng Zhou, Wending Zhao, Jianqin Yin
Publication venue and year	Arxiv	Publication URL:	https://arxiv.org/abs/2303.07141
Code Language	N/A	Hardware	N/A
Code Website	N/A	Code URL	N/A

Download

Top-Down Resnet

0.45

43.342%

Anonymous Submission

Details

Name	Top-Down Resnet
Submission Date	2022-10-06 07:46:43+00:00
Abstract	A top-down pose estimation model built on top of MMPose (https://github.com/open-mmlab/mmpose). The model uses a ResNet50 backbone to predict keypoints after detecting all people in the frame. This model is trained using stitched images from the JRDB2022 training set.
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

Download

Top-Down HRNet

0.459

42.338%

Anonymous Submission

Details

Name	Top-Down HRNet
Submission Date	2022-09-30 10:13:08+00:00
Abstract	A top-down pose estimation model built on top of MMPose (https://github.com/open-mmlab/mmpose). The model uses an HRNet backbone to predict keypoints after detecting all people in the frame. This model is trained using stitched images from the JRDB2022 training set.
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

Download

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
OSPA_Pose	lower	0.0	OSPA-Pose is a pose estimation metric which directly captures the distance between two sets of poses without a thresholding parameter.
AP
AP_0.5	higher	100%	Average Precision of pose estimation with mean OKS above 0.5 [4].

Leaderboard

Instructions

2D Pose Estimation Submissions

Additional Information Used

Evaluation Measures [1]

Reference