RDDN

This repo contains the official training and testing codes for our research.

Prerequisites

• Python >= 3.7
• Torch >= 1.6.1
• Torchvision = 0.8.2
• Pillow >= 5.1.0
• Numpy >= 1.14.3

Introduction

• train.py and test.py are the entry codes for training and testing the RDDN.
• train_data.py and val_data.py are used to load the training and validation/testing datasets. residual_dense_block.py builds the RDB block.
• perceptual.py defines the network for Perceptual Loss.
• utils.py contains all corresponding utilities.
• ./checkpoints/10_[3,5,7,9]_24/best_psnr.pth is the trained weights for outdoor in SOTS from RESIDE, where 10 stand for the number of epochs and [3,5,7,9] are milestones.

Quick Start

1. Testing

$ git clone https://github.com/Haakon68/RDDN.git
$ cd RDDN

Run test.py using default hyper-parameter settings.

$ python test.py

If you want to change the default settings (e.g. modifying the batchSize since you have multiple GPUs), simply run

$ python test.py -batchSize 32

It is exactly the same way to modify any other hyper-parameters as shown above. For more details about the meaning of each hyper-parameter, please run

$ python test.py -h

2. Training

To retrain or fine-tune the RDDN, first download the OTS (for outdoor) training datasets from RESIDE.

Then, copy hazy and clear folders from downloaded OTS to ./data/train/ and ./data/test/.

If you hope to use your own training dataset, please follow the same folder structure in ./data/train/. More details can be found in train_data.py.

After putting the training dataset into the correct path, we can train the RDDN by simply running train.py using default settings.

$ python train.py

Demonstration

Only listed several examples, more results can be found in my github.

Hazy	Groundtruth	Our result