cilix-ai

cilix-ai / gs-embedding

Public

[ICLR 2026] Implementation of the paper "Learning Unified Representation of 3D Gaussian Splatting". Rethinking 3DGS representation in neural networks with SF embeddings.

cilix-ai.github.iogs-embedding-page 3d-vision 3dgs embedding-models representation-learning
38
4
100% credibility
Found Feb 03, 2026 at 19 stars 2x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

This repository implements tools for converting 3D Gaussian Splatting scenes to compact embeddings, training models on them, and visualizing results.

How It Works

1
🔍 Discover the Tool

You stumble upon this exciting project that simplifies complex 3D scenes into smart, compact forms for easier handling.

2
💻 Get It Ready

You quickly set everything up on your computer so it's all prepared for your 3D adventure.

3
📁 Load Your 3D Scene

You choose your 3D model file, and the tool gently pulls it in to start working its magic.

4
✨ Create Compact Codes

In a flash, your detailed 3D scene transforms into neat, tiny codes that hold all the important shapes and colors.

5
🔄 Switch Forms Easily

You flip back and forth between the compact codes and full 3D models whenever you need.

6
👀 View and Cluster

You gaze at beautiful pictures of your scenes and group similar parts to explore deeper.

🎉 Achieve Smarter 3D

Your 3D creations are now perfectly ready for learning, sharing, and sparking new ideas!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 38 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is gs-embedding?

This Python repo implements the ICLR 2026 paper on learning unified embeddings for 3D Gaussian Splatting (3DGS), tackling how raw Gaussian parameters—centroids, scales, rotations, opacities, and SH coeffs—are messy and non-unique for neural network training. Users get a compact, structured representation via SF embeddings that preserve geometry and color, plus CLI tools to convert PLY files to/from embeddings (e.g., `python converter.py --gaussian2emb`), train models on synthetic or ShapeSplat data, visualize renders, and cluster embeddings. Amid github iclr 2026 leak chatter and iclr reviewer leakage github buzz, it's a drop-in for 3DGS pipelines.

Why is it gaining traction?

It stands out by fixing 3DGS's parameterization woes for NN tasks like reconstruction or generation, delivering 4x faster inference checkpoints and batched conversion without rasterizer tweaks. Developers dig the quick-start CLI, pre-trained models for emb2gaussian, and utils for PLY/NPZ datasets—far simpler than hacking raw params into PointNet or MLPs. Ties into 3dgs hype and iclr 2026 openreview discussions make it a timely experiment for neural fields.

Who should use this?

3DGS researchers fine-tuning models on Gaussian primitives, NeRF devs embedding splats for downstream tasks like interpolation, or CV engineers prototyping iclr 2026-style 3D representations from PLY scans. Ideal for those hitting limits with heterogeneous Gaussians in torch pipelines, especially around iclr 2026 deadline experiments.

Verdict

Promising early code for 3DGS embeddings, but 1.0% credibility score and 28 stars signal immaturity—docs are README-focused, no tests visible. Grab pre-trained checkpoints if you're in 3dgs neural nets; train your own cautiously until iclr 2026 reviews stabilize it.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.