EndoGLSAM: Real-Time Dense Reconstruction and Tracking in Endoscopic Surgeries using Gaussian Splatting.
Kailing Wang*, Chen Yang*, Yuehao Wang, Sikuang Li, Yan Wang, Qi Dou, Xiaokang Yang, Wei Shen†

• Release reconstruction results for comparison
• Release preprocessed dataset
• Release code
• Release paper

You can install them following the instructions below.

conda create -n endogslam python=3.10 # recommended
conda activate endogslam
# torch and cuda version according to your env and device
pip install torch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

Latest version is recommended for all the packages unless specified, but make sure that your CUDA version is compatible with your pytorch.

Tested machines: Ubuntu22.04+RTX4090, Ubuntu22.04+RTX2080Ti, Windows10+RTX2080.

We use the C3VD dataset. You can use the scripts in data/prepeocess_c3vd to preprocess the dataset. We also provide the preprocessed dataset: Google Drive or My Site.

The reconstruction results for comparison is also available: Google Drive or My Site.

After you get prepared, the data structure should be like this:

- data/
  |- C3VD/
    |- cecum_t1_b/
      |- color/
      |- depth/
      |- pose.txt
    |- cecum_t3_a/
- scripts/
  |- main.py
- utils/
- other_folders/
- readme.md

If you want to use your own dataset, you can modify the dataloader or organize your data in the same structure.

Training arguments can be found in scripts/main.py. To use the default setting:

python scripts/main.py configs/c3vd/c3vd_base.py

To evaluate on a single scene:

python scripts/calc_metrics.py --gt data/C3VD/sigmoid_t3_a --render experiments/C3VD_base/sigmoid_t3_a --test_single

We use the same visualization scripts as SplaTAM for debug only.

We would like to acknowledge the following inspiring work:

• 3DGS (Bernhard Kerbl et al.)
• SplaTAM (Nikhil Keetha et al.)

If you find this code useful for your research, please use the following BibTeX entries:

@article{wang2024endogslam,
    title={EndoGSLAM: Real-Time Dense Reconstruction and Tracking in Endoscopic Surgeries using Gaussian Splatting},
    author={Kailing Wang and Chen Yang and Yuehao Wang and Sikuang Li and Yan Wang and Qi Dou and Xiaokang Yang and Wei Shen},
    journal={arXiv preprint arXiv:2403.15124},
    year={2024}
}