MVIG-SJTU / AlphaPose

Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System
http://mvig.org/research/alphapose.html
Other
8.06k stars 1.98k forks source link

Questions Regarding AlphaPose Implementation and Output Interpretation #1217

Open ignacio-barrera opened 5 months ago

ignacio-barrera commented 5 months ago

I am currently working on a project that involves testing some Human Pose Estimation Models, my principal focus its to track foots, so im using AlphaPose, OpenPose and BlazePose, because this 3 models ahs 3 foot keypoints each. So, my plan its to test this three models on COCO-WholeBody, which has 133 keypoints (3 for each feet). I have encountered a few questions and problems that would greatly appreciate your guidance, my main three problems and questions are:

All of this its using FastPose Model for COCO WholeBody dataset 133 keypoints [2] (checkpoint named coco_wholebody133_fast50_regression_256x192.pth) and this cfg file [3] (cfg file named: 256x192_res50_lr1e-3_2x-regression.yaml).

And last, the output for my inference (COCO eval2017) looks like the .json file that i adjunted.

I greatly appreciate your time and assistance in addressing these questions.

Links: https://github.com/MVIG-SJTU/AlphaPose/blob/master/scripts/validate.py https://github.com/MVIG-SJTU/AlphaPose/blob/master/docs/MODEL_ZOO.md https://github.com/MVIG-SJTU/AlphaPose/blob/master/configs/coco_wholebody/resnet/256x192_res50_lr1e-3_2x-regression.yaml

first_res.json