Deformable Kernel Network for Joint Image Filtering

This is the pytorch implementation of our DKN [Paper]

1. Requirement

Pytorch 1.0.0

Cuda 10.0

Opencv 3.4

tqdm

logging

2. Dataset

NYU v2 dataset, our split can be downloaded: http://gofile.me/3G5St/2lFq5R3TL

3. Inference

• DKN

  python inference.py --rgb images/0_rgb.png --depth images/0_lr.png --k 3 --d 15 --scale 8 --parameter parameter/DKN_8x --output images/result_dkn.png --model DKN

• FDKN

  python inference.py --rgb images/0_rgb.png --depth images/0_lr.png --k 3 --d 15 --scale 8 --parameter parameter/FDKN_8x --output images/result_fdkn.png --model FDKN

4. Train

• DKN

  python train.py --k 3 --d 15 --scale 8 --model DKN

• FDKN

  python train.py --k 3 --d 15 --scale 8 --model FDKN