search

Gengzigang / PCT

This is an official implementation of our CVPR 2023 paper "Human Pose as Compositional Tokens" (https://arxiv.org/pdf/2303.11638.pdf)
MIT License
332 stars 21 forks source link

How to inference 3D pose? #11

Open jun297 opened 1 year ago

jun297 commented 1 year ago

Hi, nice work. Thank you for sharing code, I have ran the demo, but it produces only 2D pose. I am trying to get wholebody (e.g., COCO-wholebody) 3D pose How can I get wholebody or 3D pose?

Thank you

Vivien888 commented 1 year ago

I have the same doubt, hope to get an answer, thank you

bibibabibo26 commented 1 year ago

I have the same doubt, hope to get an answer, really thanks for that

cjerry1243 commented 1 year ago

Same here. I am wondering if 3D demo/models/pipeline will be released. @Gengzigang

imabackstabber commented 1 year ago

idk, but i trained on coco eft dataset, classifier won't be able to converge... I think the reason is that it's rather hard for classifier to classify in-the-wild images to discrete vqvae entry, and now I kinda believe that PCT will only be able to handle with simple 3D dataset like human3.6m...? Any thoughts? @Gengzigang

KunmingS commented 3 months ago

I also met the problem that the classifier didn't converge. It achieved around 9 and didn't decrease. But the performance is pretty good when I try it.