GaussianImage++: Boosted Image Representation and Compression with 2D Gaussian Splatting

[paper][code]

Tiantian Li, Xinjie Zhang, Xingtong Ge, Tongda Xu, Dailan He, Jun Zhang, Yan Wang📧

(📧 denotes corresponding author.)

This is the official implementation of our paper GaussianImage++, accepted by AAAI 2026.

• 2025/12/23: 🔥 We release our Python and CUDA code for GaussianImage++ presented in our paper. Have a try!

• 2025/11/8: 🌟 Our paper has been accepted by AAAI 2026! 🎉 Cheers!

The repository contains submodules, thus please check it out with

# SSH
git clone git@github.com:Sweethyh/GaussianImage_plus.git 

or

# HTTPS
git clone https://github.com/Sweethyh/GaussianImage_plus.git 

After cloning the repository, you can follow these steps to train GaussianImage++ models under different tasks.

cd gsplat
pip install .[dev]
cd ../
pip install -r requirements.txt

If you encounter errors while installing the packages listed in requirements.txt, you can try installing each Python package individually using the pip command.

Before training, you need to download the kodak and DIV2K-validation datasets. The dataset folder is organized as follows.

├── dataset
│   | kodak 
│     ├── kodim01.png
│     ├── kodim02.png 
│     ├── ...
│   | DIV2K_valid_HR
│     ├── 0801.png
│     ├── 0802.png
│     ├── 0803.png
│     ├── ...

python train.py --num_points 2500 --max_num_points 5000 --data_name kodak -d ./dataset/kodak/

python train.py --num_points 2500 --max_num_points 5000 --data_name kodak -d ./dataset/kodak/ --color_norm 

python train_quantize.py --num_points 2500 --max_num_points 5000 --data_name kodak -d ./dataset/kodak/ --color_norm 

Our code was developed based on GaussianImage. We thank them for providing the novel framework to implement image representation and compression.