lllyasviel / lllyasviel/LayerDiffuse

Tried, Reproducing the model from the paper, but found poor quality

Open
#35 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
No language data
Stars
2.2k
Forks
36
PR merge metrics
No merged PRs in 30d

Description

hi @lllyasviel amazing work, i tried replicating the training strategy in the paper, and was able to get a model working, but the results i see are cloudy and have artifacts sometimes, lack consistency across seeds. did you regularize using any additional losses other than regular diffusion objective in latent space while training the Unet?

did you use any additional tricks not mentioned in paper?

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 comparing the reported training strategy with the paper and the repository's model-training entry points. Investigate the cloudy artifacts and seed inconsistency, including whether additional losses or undocumented training steps are involved. The issue does not define a concrete code change or a clear completion test.

Written by the indexing model from the issue text.

Assessment

Domain
machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.