This repository reimplements the training pipeline of CrossNet and provides an unoffical implemenation of CrossNet++.

The offical implemenation of CrossNet: ECCV2018_CrossNet_RefSR

Reference Papers:

1. CrossNet: An End-to-end Reference-based Super Resolution Network using Cross-scale Warping
2. CrossNet++: Cross-scale Large-parallax Warping for Reference-based Super-resolution

• PyTorch
• Python3 (tested on Python3.7)
• torchlight

Download the original light field datasets, Flower, LFVideo.

python run.py train -s saved/flower/crossnet/base -c configs/flower/crossnet.yaml
python run.py train -s saved/flower/crossnet++/base -c configs/flower/crossnet++.yaml

python run.py test -s saved/flower/crossnet/base -r best
python run.py test -s saved/flower/crossnet++/base -r best

• Serpate model for each viewpoint. Reference image are at (0,0).
• Charbonnier loss only

• EDSR