COCO 2019 DensePose Challenge

Organized by vkhalidov - Current server time: Dec. 5, 2019, 4:49 p.m. UTC

Previous

test-2019
Aug. 1, 2019, midnight UTC

Current

val
Aug. 1, 2019, midnight UTC

End

Competition Ends
Oct. 5, 2019, 6:59 a.m. UTC

COCO 2019 DensePose Challenge

The COCO DensePose Challenge is designed to push the state of the art in dense human pose estimation. The task of dense human pose estimation comprises

  • human body detection,
  • human body segmentation,
  • mapping all image pixels that belong to a human body to the 3D surface of the body.

Schematic representation of the task is given in Figure 1.

 


Figure 1. Dense human pose estimation involves human body detection, human body segmentation and mapping all pixels that belong to a human body to the 3D surface of the body.

This challenge is part of the COCO and Mapillary Workshop at ICCV 2019. For further details about the joint workshop please visit the workshop website. Please also take a look at the concurrent COCO 2019 Detection, Panoptic and Keypoints challenges.

To participate, you can find instructions on the DensePose page. In particular, please see the overview, download, data format, results format, upload and evaluation pages for more detail.

Important: please note the change in evaluation metric (see evaluation)! All baselines on the leaderboard have been reevaluated using the new metric.

The COCO DensePose API is used to evaluate results of the DensePose Challenge. The software uses both candidate and reference point correspondences defined by body part index and intrinsic 2D coordinates on the body part, and applies the Average Precision evaluation metric. For more details please check evaluation procedure, results format and upload instructions pages.

Important: please note the change in evaluation metric (see evaluation)! All baselines on the leaderboard have been reevaluated using the new metric.

Please refer to COCO Terms of Use.

val

Start: Aug. 1, 2019, midnight

Description: The val evaluation server for 2019 COCO DensePose challenge. We encourage use of val to perform validation experiments; for publication, please evaluate your results on test. Results submitted to val will NOT be posted to the public leaderboard on cocodataset.org. JSON file with the results should be named densepose_val_[alg]_results.json, where [alg] with your algorithm name; it should be compressed into a zip file named densepose_val_[alg]_results.zip.

test-2019

Start: Aug. 1, 2019, midnight

Description: The test evaluation server for 2019 COCO DensePose challenge. We encourage use of test to report evaluation results for publication. You can access the latest public results for comparison at http://cocodataset.org/#densepose-leaderboard. We will migrate results submitted to test regularly to the public leaderboard on cocodataset.org. Please choose "Submit to Leaderboard" if you want your submission to be appeared on our leaderboard. JSON file with the results should be named densepose_test_[alg]_results.json, where [alg] with your algorithm name; it should be compressed into a zip file named densepose_test_[alg]_results.zip.

Competition Ends

Oct. 5, 2019, 6:59 a.m.

You must be logged in to participate in competitions.

Sign In
# Username Score
1 wangxuanhan 0.65
2 fyhtea 0.61
3 zhuxuhan 0.56