Awesome

Taming 3DGS: High-Quality Radiance Fields with Limited Resources

Saswat Subhajyoti Mallick*, Rahul Goel*, Bernhard Kerbl, Francisco Vicente Carrasco, Markus Steinberger, Fernando De La Torre (* indicates equal contribution)

TL;DR We improve the densification process to make the primitive count deterministic and implement several low-level optimizations for fast convergence.

Usage

Clone the repository

git clone https://github.com/humansensinglab/taming-3dgs.git --recursive

Follow the instructions in the original 3DGS repository to setup the environment.

--cams

Number of cameras required to compute gaussian scores. Default set to 10.

--budget

The final number of gaussians to end up with. Can be a float or an integer based on --mode.

--mode

multiplier: the final count of gaussians will be multiplier x the initial (SfM) count <br> final_count: the final count of gaussians will be set exactly to final_count.

--websockets

Whether to use the web based viewer or not.

--ho_iteration

High opacity gaussians will be enabled from which iteration. Defaults to 15000 (after densification ends).

--sh_lower

Whether to enable less-frequent (once every 16 iterations) SH updates to gain speed.

--benchmark_dir

The location of the folder where the timing results are stored. No time profiling will be done if left blank.

To reproduce the scores in our paper, please run train.sh

Web viewer

We provide a basic browser-based renderer to track the training progress. Steps for usage are as follows: <br>

1. Move the web-viewer folder into the host machine from where you wish to view the training.
2. Assign a port number while spawning a training session using train.py --port <port_num> --websockets. If you're running this on a remote server, forward the port to your host system.
3. Maintain the same port number in the app.js file.
4. Open render.html in your browser.

Note

The performance optimizations that are drop-in replacements for the original implementation are available under the rasterizer branch. All the performance optimizations are released under the MIT License. Please refer to the Inria repository for complete instructions.

They have also been integrated with the original Inria repository so it can be directly used there.

Citation

If you find this repo useful, please cite:

@inproceedings{10.1145/3680528.3687694,
    author = {Mallick, Saswat Subhajyoti and Goel, Rahul and Kerbl, Bernhard and Steinberger, Markus and Carrasco, Francisco Vicente and De La Torre, Fernando},
    title = {Taming 3DGS: High-Quality Radiance Fields with Limited Resources},
    year = {2024},
    isbn = {9798400711312},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3680528.3687694},
    doi = {10.1145/3680528.3687694},
    booktitle = {SIGGRAPH Asia 2024 Conference Papers},
    articleno = {2},
    numpages = {11},
    keywords = {Radiance Fields, Gaussian Splatting},
    series = {SA '24}
}