search

shunsukesaito / PIFu

This repository contains the code for the paper "PIFu: Pixel-Aligned Implicit Function for High-Resolution Clothed Human Digitization"
https://shunsukesaito.github.io/PIFu/
Other
1.75k stars 342 forks source link

How to train model in the case of having Buff dataset #80

Closed sunjc0306 closed 3 years ago

sunjc0306 commented 3 years ago

Hi, I am very interested in your research, especially this code about PiFu model. I have gotten the Buff dataset now. How isThe Buff dataset integrated your code in the TrainDataset.py file? In additional, how are the point-to-surface Euclidean distance P2S , the Chamfer distance and the normal reprojection error calculated. what their formulas are.

shunsukesaito commented 3 years ago

You have to modify the code to incorporate BUFF dataset in the training. To do so, you have to first render BUFF dataset by modifying the released rendering code and somehow fix holes in the mesh (note that BUFF meshes are not watertight but PIFu training requires watertight mesh for ground-truth occupancy labels.) Regarding evaluation, you can look at the previous issue here. https://github.com/shunsukesaito/PIFu/issues/56