The official implementation of SAGA (Segment Any 3D GAussians)
Note: This repository is still under construction.
The official implementation of SAGA (Segment Any 3D GAussians). The paper is at this url. Please refer to our project page for more information.
SAGA can perform fine-grained interactive segmentation for 3D Gaussians within milliseconds.
The installation of SAGA is similar to 3D Gaussian Splatting.
git clone [email protected]:Jumpat/SegAnyGAussians.git
or
git clone https://github.com/Jumpat/SegAnyGAussians.git
Then install the dependencies:
conda env create --file environment.yml
conda activate gaussian_splatting
In default, we use the public ViT-H model for SAM. You can download the pre-trained model from here and put it under ./dependencies/sam_ckpt.
The used datasets are 360_v2, nerf_llff_data and LERF.
The data structure of SAGA is shown as follows:
./data
/360_v2
/garden
/images
/images_2
/images_4
/images_8
/sparse
/features
/sam_masks
...
/nerf_llff_data
/fern
/images
/poses_bounds.npy
/sparse
/features
/sam_masks
/horns
...
...
/lerf_data
...
To get the features and sam_masks, run the following command:
python extract_features.py --image_root <path to the scene data> --sam_checkpoint_path <path to the pre-trained SAM model> --downsample <1/2/4/8>
python extract_segment_everything_masks.py --image_root <path to the scene data> --sam_checkpoint_path <path to the pre-trained SAM model> --downsample <1/2/4/8>
Note that sometimes the downsample is essential due to the limited GPU memory.
We inherit all attributes from 3DGS, more information about training the Gaussians can be found in their repo.
python train_scene.py -s <path to COLMAP or NeRF Synthetic dataset>
python train_contrastive_feature.py -m <path to the pre-trained 3DGS model>
Temporarily, 3D segmentation requires jupyter notebook. Please refer to prompt_segmenting.ipynb for instructions.
After running the scripts in prompt_segmenting.ipynb, the bitmap of the Gaussians will be saved as ./segmentation_res/final_mask.pt (you can change it so that the segmentation results of many different objects can be saved). To render the segmentation results (get the segmented object), run the following command:
python render.py -m <path to the pre-trained 3DGS model> --precomputed_mask <path to the segmentation results> --target scene --segment
To get the 2D rendered masks, run the following command:
python render.py -m <path to the pre-trained 3DGS model> --precomputed_mask <path to the segmentation results> --target seg
You can also render the pre-trained 3DGS model without segmentation:
python render.py -m <path to the pre-trained 3DGS model> --target scene
If you find this project helpful for your research, please consider citing the report and giving a ⭐.
@article{cen2023saga,
title={Segment Any 3D Gaussians},
author={Jiazhong Cen and Jiemin Fang and Chen Yang and Lingxi Xie and Xiaopeng Zhang and Wei Shen and Qi Tian},
year={2023},
journal={arXiv preprint arXiv:2312.00860},
}