Skip to content

Latest commit

 

History

History
46 lines (31 loc) · 2.5 KB

installation.md

File metadata and controls

46 lines (31 loc) · 2.5 KB

Installation

The code is tested with Python == 3.8, Pytorch == 1.13.1, CUDA == 11.6, mmengine == 0.10.3, mmdet3d == 1.4.0, mmcv == 2.0.0, mmdet == 3.2.0 and MinkowskiEngine == 0.5.4. We recommend you to use Anaconda to make sure that all dependencies are installed correctly.

Step 1: Download and install Miniconda from the official website.

Step 2: Create a new conda environment and activate it:

conda create -n ESAM python=3.8
conda activate ESAM

Step 3: Install PyTorch following official instructions, e.g.

conda install pytorch torchvision -c pytorch

Step 4: Follow mmdetection3d to install mmcv, mmdet3d and mmdet.

Step 5: Install MinkowskiEngine:

conda install openblas-devel -c anaconda
pip install -U git+https://github.com/NVIDIA/MinkowskiEngine -v --no-deps --install-option="--blas_include_dirs=/opt/conda/include" --install-option="--blas=openblas"

Step 6: Install SAM & FastSAM:

  • Please follow here for installation of SAM. Then download the checkpoint for Vit-H SAM model and put it in the folder 'data'.

  • Please follow here for installation of FastSAM. Then download the checkpoint for FastSAM and put it in the folder 'data'.

Step 7: Download backbone checkpoint:

We follow Oneformer3D to initialize the backbone from Mask3D checkpoint. It should be downloaded and put to work_dirs/tmp before training.

Step 8: Follow SAM3D to install pointops.

Step 9: Get segmentator repository:

Please refer to segmentator to get mesh segmentator. You should clone the repository in the folder 'data', and it will be imported in 'batch_load_scannet_data.py' for generating mesh segmentation results.

Step 10: Install other dependencies:

pip install -r requirements.txt