Fast 2DGS: Efficient Image Representation with Deep Gaussian Prior

As generative models become increasingly capable of producing high-fidelity visual content, the demand for efficient, interpretable, and editable image representations has grown substantially. Recent advances in 2D Gaussian Splatting (2DGS) have emerged as a promising solution, offering explicit control, high interpretability, and real-time rendering capabilities ($>1000$ FPS). However, high-quality 2DGS typically requires post-optimization. Existing methods adopt random or heuristics (e.g., gradient maps), which are often insensitive to image complexity and lead to slow convergence ($>10s$). More recent approaches introduce learnable networks to predict initial Gaussian configurations, but at the cost of increased computational and architectural complexity. To bridge this gap, we present \textbf{Fast-2DGS}, a lightweight framework for efficient Gaussian image representation. Specifically, we introduce Deep Gaussian Prior, implemented as a conditional network to capture the spatial distribution of Gaussian primitives under different complexities. In addition, we propose an attribute regression network to predict dense Gaussian properties. Experiments demonstrate that this disentangled architecture achieves high-quality reconstruction in a single forward pass, followed by minimal fine-tuning. More importantly, our approach significantly reduces computational cost without compromising visual quality, bringing 2DGS closer to industry-ready deployment.

(will update more)

Setup

1. Navigate to projekt folder, Create a new Python environment and install the dependencies (double check CUDA path). We use torch==2.7.1+cu126 for this study, since torch==2.9 found very slow on gsplat.

   conda create -n 2dgs python=3.12
   conda activate 2dgs
   pip install -r requirements.txt
   git clone https://github.com/Aztech-Lab/gmod.git
   cd gmod
   pip install -e . --no-build-isolation
   cd ..

*Note: You might encounter CUDA path issue when install gmod, set CUDA path to match your torch version, then install gmod again.
*The gmod package is build by Image-GS team. We rename it to gmod to aviod conflict with other gsplat library.

Now you can run python test_2dgs.py for a simple demo of 2DGS on Bulbasaur, results are saved in ./test/fit_outputs.

2. Clone the datasets from 2DGS_dataset, download DIV2K_train_HR and organize the folder structure as follows:

   git clone https://github.com/Aztech-Lab/2DGS_dataset.git

   2DGS_dataset
   └── dataset
       ├── Kodak
       └── DIV2K
           └── DIV2K_train_HR (need to download)
           └── DIV2K_valid_HR
       └── ImageGS_anime
       └── ImageGS_textures
   

Training

(will update soon)

Testing

(will update soon)

Framework

Results

Gaussian Initialization Comparison

Impact of Gaussians

Additional Results

Acknowledgements

We sincerely appreciate the Image-GS team for providing the 2DGS rendering core and for sharing their high-quality datasets, and we thank Instant-GI team for their great work and deep inspiration. Moreover, we thank the GaussianImage team for their foundation work at this domain.

Citation

If you find this project helpful to your research, please consider citing BibTeX:

@article{wang2025fast,
  title={Fast 2DGS: Efficient Image Representation with Deep Gaussian Prior},
  author={Wang, Hao and Bastola, Ashish and Zhou, Chaoyi and Zhu, Wenhui and Chen, Xiwen and Dong, Xuanzhao and Huang, Siyu and Razi, Abolfazl},
  journal={arXiv preprint arXiv:2512.12774},
  year={2025}
}