DRRN-pytorch

This is an unofficial implementation of "Deep Recursive Residual Network for Super Resolution (DRRN)", CVPR 2017 in Pytorch. [Paper]

You can get the official Caffe implementation here.

Usage

Training

usage: main.py [-h] [--batchSize BATCHSIZE] [--nEpochs NEPOCHS] [--lr LR]
               [--step STEP] [--cuda] [--resume RESUME]
               [--start-epoch START_EPOCH] [--clip CLIP] [--threads THREADS]
               [--momentum MOMENTUM] [--weight-decay WEIGHT_DECAY]
               [--pretrained PRETRAINED]
               
optional arguments:
  -h, --help            Show this help message and exit
  --batchSize           Training batch size
  --nEpochs             Number of epochs to train for
  --lr                  Learning rate. Default=0.1
  --step                Learning rate decay, Default: n=5 epochs
  --cuda                Use cuda?
  --resume              Path to checkpoint
  --clip                Clipping Gradients. Default=0.01
  --threads             Number of threads for data loader to use Default=1
  --momentum            Momentum, Default: 0.9
  --weight-decay        Weight decay, Default: 1e-4
  --pretrained          Path to the pretrained model, used for weight initialization (default: none)

Evaluation

usage: eval.py [-h] [--cuda] [--model MODEL] [--dataset DATASET]
               [--scale SCALE]

PyTorch DRRN Evaluation

optional arguments:
  -h, --help         show this help message and exit
  --cuda             use cuda?
  --model MODEL      model path
  --dataset DATASET  dataset name, Default: Set5

An example of training usage is shown as follows:

python eval.py --cuda

Prepare Training dataset

the training data is generated with Matlab Bicubic Interpolation, please refer Code for Data Generation for creating training files.

Performance

We provide a rough pre-trained DRRN_B1U25 model trained on 291 images with data augmentation. The model can achieve a better performance with a smart optimization strategy. For the DRRN_B1U9 implementation, you can manually modify the number of recursive blocks here.
The same adjustable gradient clipping's implementation as original paper.
No bias is used in this implementation.
No batch normalization is used in this implementation.
Performance in PSNR on Set5