# Create conda env
conda create -n gcn_of python=3.9
conda activate gcn_of
# Install PyTorch (CUDA 12.1/12.8 depending on your setup; adjust if needed)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
# or:
# pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
# Install Python dependencies
pip install \
omegaconf \
opencv-python \
matplotlib \
psutil \
wandb \
lightning \
numba \
pybind11 \
tqdm \
pandas \
loguru \
pycocotools
# Install HDF5 related libs
conda install h5py
conda install -c conda-forge blosc-hdf5-plugin
# Inside main folder
cd GCN-OF
python setup.py build_ext --inplaceOptical flow GT:
- Download MVSEC optical flow ground truth from: https://daniilidis-group.github.io/mvsec/download/
Pre-processed MVSEC data:
- Additional data from: https://drive.google.com/drive/folders/1rwyRk26wtWeRgrAx_fgPc-ubUzTFThkV
Convert .npz β .npy:
Use convert_npz_to_npy.py inside dataset/utils to convert the original MVSEC .npz flow files into .npy:
python dataset/utils/convert_npz_to_npy.pyYour data folder should look like this:
data/
βββ mvsec
βββ indoor_flying
βββ indoor_flying1_data.hdf5
βββ indoor_flying1_gt_flow_dist
β βββ timestamps.npy
β βββ x_flow_dist.npy
β βββ y_flow_dist.npy
βββ indoor_flying1_gt_flow_dist.npz
βββ indoor_flying2_data.hdf5
βββ indoor_flying2_gt_flow_dist
β βββ timestamps.npy
β βββ x_flow_dist.npy
β βββ y_flow_dist.npy
βββ indoor_flying2_gt_flow_dist.npz
βββ indoor_flying3_data.hdf5
βββ indoor_flying3_gt_flow_dist
β βββ timestamps.npy
β βββ x_flow_dist.npy
β βββ y_flow_dist.npy
βββ indoor_flying3_gt_flow_dist.npz
βββ indoor_flying4_data.hdf5
βββ indoor_flying4_gt_flow_dist
β βββ timestamps.npy
β βββ x_flow_dist.npy
β βββ y_flow_dist.npy
βββ indoor_flying4_gt_flow_dist.npz