[CVPR 2026] SGI

Official Pytorch implementation of SGI: Structured 2D Gaussians for Efficient and Compact Large Image Representation.

Zixuan Pan*, Kaiyuan Tang*, Jun Xia, Yifan Qin, Lin Gu, Chaoli Wang, Jianxu Chen, Yiyu Shi

(* denotes equal contribution)

Overview

2D Gaussian Splatting has emerged as a novel image representation technique that can support efficient rendering on low-end devices. However, scaling to high-resolution images requires optimizing and storing millions of unstructured Gaussian primitives independently, leading to slow convergence and redundant parameters. To address this, we propose Structured Gaussian Image (SGI), a compact and efficient framework for representing high-resolution images. SGI decomposes a complex image into multi-scale local spaces defined by a set of seeds. Each seed corresponds to a spatially coherent region and, together with lightweight multi-layer perceptrons (MLPs), generates structured implicit 2D neural Gaussians. This seed-based formulation imposes structural regularity on otherwise unstructured Gaussian primitives, which facilitates entropy-based compression at the seed level to reduce the total storage. However, optimizing seed parameters directly on high-resolution images is a challenging and non-trivial task. Therefore, we designed a multi-scale fitting strategy that refines the seed representation in a coarse-to-fine manner, substantially accelerating convergence. Quantitative and qualitative evaluations demonstrate that SGI achieves up to 7.5 $\times$ compression over prior non-quantized 2D Gaussian methods and 1.6 $\times$ over quantized ones, while also delivering 1.6 $\times$ and 6.5 $\times$ faster optimization, respectively, without degrading, and often improving, image fidelity.

Performance

Installation

We tested our code on a server with Ubuntu 20.04.1, cuda 11.8, gcc 9.4.0.

1. Unzip files

cd submodules
unzip diff-gaussian-rasterization.zip
unzip gridencoder.zip
unzip simple-knn.zip
unzip arithmetic.zip
cd ..

2. Install environment

conda env create --file environment.yml
conda activate SGI_env

3. Install gsplat2d

cd gsplat2d/gsplat2d/cuda/csrc
mkdir third_party
cd third_party
git clone https://github.com/g-truc/glm.git
cd ../../../..
python setup.py build
python setup.py install

4. Install tmc3 (for GPCC)

• Please refer to tmc3 github for installation.
• Don't forget to add tmc3 to your environment variable.
• Tips: tmc3 is commonly located at /PATH/TO/mpeg-pcc-tmc13/build/tmc3.

Data

The data structure should be organised as follows:

data/
├── dataset_name
│   ├── xxx_0.png
│   ├── xxx_1.png
│   ├── xxx_2.png
│   ├── ...
...

Public Data

• The FGF2 dataset can be downloaded here
• The ICB dataset can be downloaded here
• The STimage datasets can be downloaded here

Training

Set the path of tmc3 before running:

bash train.sh image_dir output_root

Notes:

• The pipeline runs training, encoding, decoding, and evaluation.
• Outputs are organized as <output_root>/<dataset_name>/<image_name>/....
• Logs are written to outputs.log under each image's output directory.
• Encoded bitstreams are saved to <model_path>/bitstreams.
• Per-image metrics are saved to <output_root>/<dataset_name>/metrics.json and <output_root>/<dataset_name>/metrics.csv.
• For very large images, use --disable_lpips to skip LPIPS and avoid potential GPU OOM during evaluation.

Acknowledgement

This codebase is built upon several excellent open-source projects, including LIG, GaussianImage, gsplat, and HAC-plus. We sincerely thank the authors of these works for making their code publicly available.

Citation

If you use SGI algorithm in your research, please cite our paper:

@misc{pan2026sgistructured2dgaussians,
      title={SGI: Structured 2D Gaussians for Efficient and Compact Large Image Representation}, 
      author={Zixuan Pan and Kaiyuan Tang and Jun Xia and Yifan Qin and Lin Gu and Chaoli Wang and Jianxu Chen and Yiyu Shi},
      year={2026},
      eprint={2603.07789},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.07789}, 
}