chungyiweng / humannerf

HumanNeRF turns a monocular video of moving people into a 360 free-viewpoint video.
MIT License
786 stars 86 forks source link

Visualization of skinning weight #36

Open kwea123 opened 2 years ago

kwea123 commented 2 years ago

Have you ever visualized the skinning weight produced by MotionWeightVolumeDecoder? I would like to compare to other methods like the ones you mention in the paper. Basically I expect something like (taken from here)

スクリーンショット 2022-08-31 12 21 04

in order to make sure the volume is assigned to the correct joint

SDNAFIO commented 2 years ago

Hi, I visualized the skinning weights, and the result does look quite off from what one would expect. However, consider that I'm not one of the authors, so take this with a grain of salt. On the left, I plotted the weights as they are initialized (motion_weights_priors), and on the right, I plotted the learned weights. Data is taken from the pretrained model for zju mocap subject 386.

skinning_weights

Andyen512 commented 2 years ago

@SDNAFIO Hi, can you share the weight visualization code? I'd like to see my trained model.

Jia-Wei-Liu commented 2 years ago

@SDNAFIO Hi, Thanks for sharing your result! If possible, could you please share your code for visualising the skinning weights? Thanks much!