microsoft / microsoft/TRELLIS

VAE mesh decoder training data

Open
#236 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.7k
Forks
1.3k
PR merge metrics
No merged PRs in 30d

Description

Hello, I noticed that you use normal map to calculate color loss as follows:

a normal map N_m directly derived from the mesh,

However, I try to find the corresponding code here, and I find you didn't use the rendered mesh gt normal map in structured_latent_vae_mesh_dec.py:line 213, but instead use normal map from dataloader, where it doesn't provide any normal map data. So, I wonder if you indeed not use normal map recon loss in color loss. Thank you for your reply!
https://github.com/microsoft/TRELLIS/blob/eb83038919f6e1feb63accf3a97a377a608c497d/trellis/trainers/vae/structured_latent_vae_mesh_dec.py#L171-L216
https://github.com/microsoft/TRELLIS/blob/eb83038919f6e1feb63accf3a97a377a608c497d/trellis/datasets/structured_latent2render.py#L90-L112

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reading trellis/trainers/vae/structured_latent_vae_mesh_dec.py around line 213 and compare it with trellis/datasets/structured_latent2render.py around lines 90-112. Trace whether the dataloader supplies the normal-map data referenced in the issue and whether the color-loss path consumes it. Done means the discrepancy is confirmed or disproved with a documented explanation and, if confirmed, a clearly scoped fix.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-graphics, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.