You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
Training NeRFs often results in common rendering artifacts (e.g., floaters or noisy regions) that require manual intervention. This can be time-consuming and inefficient, especially for users who want quick visual results.
Describe the solution you'd like
I’ve implemented a simple post-processing method that cleans up NeRF artifacts in about 30 seconds per scene. It can be used as a final step after training, improving the visual quality without needing to retrain. I’d like to contribute this method either as an external plugin or directly into the main Nerfstudio repository, depending on what the team prefers.
Describe alternatives you've considered
Keeping it as a standalone external method hosted on my GitHub, with documentation for users to integrate manually.
Exploring integration via the existing Nerfstudio method registry.
Using training-time regularization instead of postprocessing, though it’s slower and less flexible.
Additional context
The method is already functional and tested on several scenes, datasets and NeRF methods. It’s designed to be fast and easy to use. Here’s a link to the GitHub repo with code and example outputs: https://leosegre.github.io/optimize-the-unseen/ Happy to provide more technical details or adapt it for better integration!
The text was updated successfully, but these errors were encountered:
I’ve built a simple post-processing method to clean up NeRF artifacts—repo is here: https://leosegre.github.io/optimize-the-unseen/
Is your feature request related to a problem? Please describe.
Training NeRFs often results in common rendering artifacts (e.g., floaters or noisy regions) that require manual intervention. This can be time-consuming and inefficient, especially for users who want quick visual results.
Describe the solution you'd like
I’ve implemented a simple post-processing method that cleans up NeRF artifacts in about 30 seconds per scene. It can be used as a final step after training, improving the visual quality without needing to retrain. I’d like to contribute this method either as an external plugin or directly into the main Nerfstudio repository, depending on what the team prefers.
Describe alternatives you've considered
Additional context
The method is already functional and tested on several scenes, datasets and NeRF methods. It’s designed to be fast and easy to use. Here’s a link to the GitHub repo with code and example outputs: https://leosegre.github.io/optimize-the-unseen/
Happy to provide more technical details or adapt it for better integration!
The text was updated successfully, but these errors were encountered: