8000 GitHub - Slian22/CompGS-GSPLaT: An optimized fork of Compact3D (CompGS: ECCV 2024), replacing the original rasterizer with GSPLaT for improved compatibility, speed, and gradient support in 3D Gaussian Splatting training and rendering pipelines.
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

An optimized fork of Compact3D (CompGS: ECCV 2024), replacing the original rasterizer with GSPLaT for improved compatibility, speed, and gradient support in 3D Gaussian Splatting training and rendering pipelines.

License

Notifications You must be signed in to change notification settings

Slian22/CompGS-GSPLaT

Repository files navigation

CompGS-GSPLaT: Compact3D with GSPLaT Rasterizer

This repository is a fork of Compact3D (CompGS), accepted to ECCV 2024, where the original diff-gaussian-rasterization backend has been fully replaced by Nerfstudio's GSPLaT rasterizer.

We preserve the full training and evaluation pipeline of Compact3D while leveraging GSPLaT’s modern, CUDA-efficient rasterization backend, which provides:

  • ~20% training speedup
  • Lower memory usage
  • Slight improvements in rendering quality
  • Compatibility with OpenGL-style camera & gradient flow

Overview

Compact3D is a method to compress large 3D Gaussian Splatting models via vector quantization of Gaussian parameters (color & covariance).
This fork integrates the GSPLaT rasterizer, yielding faster and more memory-efficient training, especially on large scenes.

Installation (GSPLaT Backend)

We use the GSPLaT rasterizer from nerfstudio-project/gsplat.

Please follow the official installation instructions from GSPLaT to set up the rasterization backend and its dependencies:
👉 https://github.com/nerfstudio-project/gsplat#installation

Once installed, make sure the gsplat Python package is available in your environment.

Getting Started

Please follow the official Compact3D documentation for data preparation, training, rendering, and evaluation:
👉 https://github.com/UCDvision/compact3d#readme

Modifications Summary

The following components have been modified or extended to support gsplat rendering:

- scene/gaussian_model.py

→ Rewrote add_densification_stats() to support absgrad and screen-space scaling based on image size.

- gaussian_renderer/__init__.py

→ Replaced all rasterizer calls with gsplat.rasterization().

- train_kmeans.py

→ Updated training loop to support GSPLaT gradients and max_radii2D image-space scaling.

→ Added memory usage printout (torch.cuda.max_memory_allocated()).

These changes are minimal, focused, and preserve the original Compact3D pipeline and structure.

Citation

If you use this repository, please cite:

@article{navaneet2023compact3d,
  title={CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization},
  author={Navaneet, KL and Meibodi, Kossar Pourahmadi and Koohpayegani, Soroush Abbasi and Pirsiavash, Hamed},
  journal={ECCV},
  year={2024}
}

About

An optimized fork of Compact3D (CompGS: ECCV 2024), replacing the original rasterizer with GSPLaT for improved compatibility, speed, and gradient support in 3D Gaussian Splatting training and rendering pipelines.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published
0