We build our code upon VA-DepthNet.
We recommend you to set up a conda environment with the requried packages.
conda env create -f environment.yml
Download the pretrained encoder backbone from here. The code currently supports Swin-T, Swin-S, Swin-L backbones.
Training VA-DepthNet model for NYUv2 with Swin-T backbone
python vadepthnet/train.py configs/tiny_arguments_train_nyu.txt
Training VA-DepthNet++ model for NYUv2 with Swin-T backbone with window size 2
python vadepthnet/train.py configs/window_2_arguments_train_nyu.txt
Training VA-DepthNet++ model for NYUv2 with Swin-T backbone with window size 3
python vadepthnet/train.py configs/window_3_arguments_train_nyu.txt
Training Flow Gradient model for NYUv2 with Swin-T backbone
python vadepthnet/train.py configs/flow_tiny_arguments_train_nyu.txt