ViTAE-Transformer/ViTPose

The checkpoints of "whole body" are wrong.

mornfairy opened this issue · 5 comments

The checkpoints of wholebody are wrong, the inference result is very low confidence, about 0.0000-0.00001. When I visualize the result, the key points are disordered. Can you provide the correct checkpoints of wholebody?

Hi, I got the size mismatch error. Which checkpoint do you use? @mornfairy

Hi, were you able to find out more about this? @mornfairy

@Annbless @seaman1900 @AK391 Do the wholebody models work for you?

@mornfairy @Annbless @seaman1900 @AK391

  1. Split weights with this modification

  2. Load a model with splitted weights

You can properly inference an image after the above steps with wholebody weights.

sample_ViTPose+_huge_coco_wholebody

You can easily predict an image with https://github.com/developer0hye/onepose.