I'm using the python coco API to run evaluation for Key points estimation. I have two files, a ground truth json, and a results json.

My goal is to evaluate KP estimation performance for ALL KP classes.

For example, for category 'person', I want something like this:

kp_category : 0 : 0.410733757610904   # "nose" AP
kp_category : 1 : 0.20226150054237374 # "right_eye" AP

However, the results are averaged across all key point classes.

The only thing that I found that might help is this SO question: COCO API evaluation for subset of classes. But the problem is that the suggested solution is good for outputting evaluation of different object categories and not their key points.

Any suggestions would be appreciated.