Handle alpha images properly for instant-ngp and tensorf models #2979
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Michael-Spleenlab
pushed a commit
to Michael-Spleenlab/nerfstudio
that referenced
this pull request
Apr 26, 2024
…studio-project#2979) Co-authored-by: Đinh Trinh <[email protected]>
ArpegorPSGH
pushed a commit
to ArpegorPSGH/nerfstudio
that referenced
this pull request
Jun 22, 2024
…studio-project#2979) Co-authored-by: Đinh Trinh <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
instant-ngp and tensorf models are currently not handling training on transparent images properly as other models do, resulting in worse performance on synthetic datasets.
Currently, the alpha channel is simply being removed which prevents proper handling of background rendering which
blend_background_for_loss_computationis supposed to handle for support of rendering random background color during training etc.This PR just changes the code to pass the full image to be consistent with other models:
nerfstudio/nerfstudio/models/nerfacto.py
Line 365 in 3a4ed80
nerfstudio/nerfstudio/models/mipnerf.py
Line 145 in 3a4ed80