ExGS: Extreme 3D Gaussian Compression with Diffusion Priors

🔥 News

• [2025-09] Our paper "ExGS: Extreme 3D Gaussian Compression with Diffusion Priors" is released!
• Code and pretrained models will be available soon. Stay tuned!

📖 Overview

ExGS is a novel feed-forward framework for 3D Gaussian Splatting (3DGS) compression that achieves:

• Extreme compression (over 100× reduction in storage size).
• High rendering fidelity, preserving fine geometry and texture.
• Fast inference, suitable for real-time rendering.

Qualitative comparison under different compression ratios.

🚀 Coming Soon

The official implementation (PyTorch) and pretrained models are being prepared.
We will release:

• Training code
• Inference pipeline
• Pretrained checkpoints
• Evaluation scripts

Stay tuned! ⭐ Watch this repo to get notified.