README.md

Depth Completion Pipeline

Prepare for the environment

• Install basic packages

    pip install matplotlib tensorboard scipy einops opencv-python pypng 
    pip install nuscenes-devkit scikit-image pillow

• Install mseg

    cd ./external
    pip install -e ./mseg-semantic/mseg-api
    pip install -e ./mseg-semantic

• Build flow model

    cd ./SeparableFlow-main/libs/GANet
    TORCH=$(python -c "import os; import torch; print(os.path.dirname(torch.__file__))")
    python setup.py build
    cp -r build/lib* build/lib

Download Pretrained models

• Put the pretrained models in the following path:
  1. mseg-semantic/mseg_semantic/checkpoints/mseg-3m.pth
  2. SeparableFlow-main/models/sepflow_universal.pth
  3. Sparse-Depth-Completion/Saved/model_best_epoch.pth.tar

Run the pipeline

1. Run `python scripts/run.py`
2. The output will be saved in data/scenes/depths