[WACV'26] ForestSplats: Deformable transient field for Gaussian Splatting in the Wild [ArXiv] [Project Page]

git clone https://github.com/kalelpark/ForestSplats.git --recursive

Follow the instructions in the WildGaussians to setup the environment. To download datasets and split train/test, please follow the NeRFBaselines.

python scripts.py

To understand the our superpixel-based masking strategy more easily, please refer to mask_module.py. For the uncertainty-aware density control, please also refer to submodules.

Note that the implementation includes only the masking and densification strategies.
If you have any questions or issues during the reimplementation, please feel free to reach out.

For Appearance Variation, modify the line 535 of evaluation.py.

We thank the following projects for enabling and motivating ForestSplats:

• WildGaussians
• SuperPixels
• PixelGS