Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"
Apache License 2.0
1k stars 120 forks source link

i had some weird result in rendering #112

Closed noah0924 closed 8 months ago

noah0924 commented 9 months ago

Hi, I appreciate for your great contributions.

I followed the colab version of #68

This is my 2D prediction.

image

But 3D rendering result is this.

image

Some keypoints are fixed on the left top location

Here is another one for another video.

image

I think 2D pose doesn't have any problem.

How could I fix this problem?

noah0924 commented 8 months ago

I found the reason for this. #68 used the alphapose of coco. It is resolved when i had changed the version of alphapose from coco to halpe26.