• Python=3.9
• pip=21.2
• pytorch=1.10
• torchvision=0.11
• cudatoolkit=11.3
• cupy=9.5
• h5py=3.3
• scikit-image=0.18
• scipy=1.7.1
• setuptools=59.5.0
• configargparse=1.5.3
• segmentation-models-pytorch=0.2
• tqdm=4.62
• tensorboard

To reproduce our results, create a data directory (e.g. ./data) with the three datasets:

• NYUv2: Download the labeled dataset from [here] and place the nyu_depth_v2_labeled.mat in ./data/NYU Depth v2.
• Middlebury: Download the 2005-2014 scenes (full size, two-view) from [here] and place the extracted scenes in ./data/Middlebury/<year>/<scene>. For the 2005 dataset, make sure to only put the scenes for which ground truth is available. The data splits are defined in code.
• DIML: Download the indoor data sample from [here] and extract it into ./data/DIML/{train,test} respectively. Then run python scripts/create_diml_npy.py ./data/DIML to create numpy binary files for faster data loading.

Our pretrained model checkpoints which were used for the numbers in our paper, for all three datasets and upsampling factors, are uploaded in the path./checkpoint of the repository.

As mentioned in the paper, the training of this model is divided into two steps:

For step 1:

python run_train.py --dataset <...> --data-dir <...> --scaling <...> --save-dir <...>

for step 2:

python run_train.py --dataset <...> --data-dir <...> --scaling <...> --save-dir <...> --step1_path <...> --step2

For test set evaluation:

python run_eval.py --checkpoint <...> --dataset <...> --data-dir <...> --scaling <...>