open-mmlab / mmpose

OpenMMLab Pose Estimation Toolbox and Benchmark.
https://mmpose.readthedocs.io/en/latest/
Apache License 2.0
5.5k stars 1.2k forks source link

Add all-in-one graphic with all methods #613

Open ioctl-user opened 3 years ago

ioctl-user commented 3 years ago

There are a lot of methods this is difficult to compare, so I have the following suggestion:

Add graphical picture for each data set with all methods with two axis: FPS and AP Something like this: https://paperswithcode.com/sota/pose-estimation-on-coco-test-dev

innerlee commented 3 years ago

Thanks for the suggestion. We have plans for fancy modelzoo visualizations (including comparisons/filtrations/sorting/etc.). We are waiting for front-end programmers 😄 If anyone one can help, then we can bring them earlier.

ioctl-user commented 3 years ago

Thanks!

As an additional suggestion: Add one more metric to existing tables, that indicate stability of the top_down methods depending on box detector quality. I.e., dependence of a given method AP quality from a box AP quality.

That may be useful for performance optimization, because one can prefer take fast box detector and steady pose detector.

innerlee commented 3 years ago

@ioctl-user Please checkout the new modelzoo!