search

tianyeli / tofu

Official code for ICCV 2021 oral paper: Topologically Consistent Multi-View Face Inference Using Volumetric Sampling
Other
149 stars 8 forks source link

Pretrain feature extractor #2

Open yjhong89 opened 2 years ago

yjhong89 commented 2 years ago

Dear author.

I am YJHong and researching on vertex refinement.

Thanks for your great works!

Since the code is not released yet, I have few questions about implementation.

  1. How pretrain feature extractor (UNet with ResNet34 backbone) ?
    • which datasets and loss function ?
    • is it freezed when training V2V ?
  2. How finetune to new datasets (case of generalization to CoMA dataset)
    • Finetune whole architecture including global and local stage ?
    • If it were true, how set ground truth vertices for each stage ? (The number of CoMA dataset vertices is different to ToFu)

Thank you! YJHong.