aigc-apps / aigc-apps/VideoX-Fun

mask_pixel_values for I2V is inconsistent during training and inference

Open
#354 5 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
2.3k
Forks
188
Avg merge
2d 22h
Merged PRs (30d)
3

Description

During training, the non-input regions are assigned a value of -1 (as seen here),
but during inference, the non-input regions are set to 0 (as shown here).

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

Compare the mask_pixel_values assignments in videox_fun/data/dataset_image_video.py around line 506 and videox_fun/pipeline/pipeline_wan_fun_inpaint.py around line 617. Trace how the masks are consumed during training and inference, then make their non-input-region values consistent. Done means both paths use the same intended mask value.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
52/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.