C2PD: Continuity-Constrained Pixelwise Deformation for Guided Depth Super-Resolution

Dependencies

python=3.9
pytorch=1.10
torchvision=0.11
cudatoolkit=11.3
segmentation-models-pytorch==0.2
scikit-image
h5py
tqdm

Dataset

1. NYU
2. Lu & Middlebury
3. RGB-D-D

test_data
└───NYU
│   └───gt
│   │   └───1001.npy
│   │   |   1002.npy
│   │   │   ...
│   │   │   1449.npy
│   └───rgb
│   │   └───1001.npy
│   │   |   1002.npy
│   │   │   ...
│   │   │   1449.npy
└───Lu
│   └───gt
│   └───rgb
└───Middlebury
│   └───gt
│   └───rgb
└───RGBDD
│   └───gt
│   └───rgb

Pretrained Models

Our pretrained model checkpoints can be downloaded here.

Training

python train.py --scale=8

Testing

Test on synthetic datasets:

python test.py --scale=8

Test on real-world RGB-D-D:

python test.py --real True