preritj / pose_estimation Goto Github PK

I tested the pretrained model using the inference.py(added padding to avoid the distortion of image) with coco val images and ai_challenge images, the result seems abnormal even in most easy cases. Only very few images works well.

Hi @preritj

Where can I acces the code?

What is the difference between keypoints, skeleton and train_keypoints and train_skeletons in the config.yaml file?
If I wish to train on COCO keypoints (including eyes and ears...) should I edit both and to edit coco.py so it doesn't have to estimate the place of the neck and head_top?