search

brjathu / PHALP

Code repository for the paper "Tracking People by Predicting 3D Appearance, Location & Pose". (CVPR 2022 Oral)
Other
259 stars 38 forks source link

Use a more complex model such as SMPL-X for hands and feet capture #24

Open cga-cmu opened 6 months ago

cga-cmu commented 6 months ago

I get the message WARNING: You are using a SMPL model, with only 10 shape coefficients.

How do I use a more complex model such as SMPL-X, which does a better job on hands and feet?

Thanks Chris Atkeson

smandava98 commented 4 months ago

Well PHALP used underlying HMR which does not predict hands and feet. I suppose you could train another model using MANO that does do hands and include coco whole body 2d keypoints model in your synthetic ground truth to re-train the HMR2 model to include foot keypoints.

I know they released HaMER (geopavlakos / hamer which does do the hands. Maybe you can combine this alongside with a re-trained HMR2 head with feet keypoints. If you manage to do this, lmk because I am interested in a fully body, accurate tracking as well.