Pytorch implement: Residual Dense Network for Image Super-Resolution
Two advantage ideas of the paper:
- join denese connect layer to ResNet
- concatenation of hierarchical features
Different with the paper, I just use there RDBs(Residual dense block), every RDB has three dense layers. So ,this is a sample implement the RDN(Residual Dense Network) proposed by the author.
- python3.5 / 3.6
- pytorch >= 0.2
- opencv
you need prepare DIV2K dataset (./data/)
train model :
python3 main.py --model_name 'RDN' --load demo_x3_RDN --dataDir ./DIV2K/ --need_patch True --patchSize 144 --nDenselayer 3 --nFeat 64 --growthRate 32 --scale 3 --epoch 10000 --lrDecay 2000 --lr 1e-4 --batchSize 16 --nThreads 4 --lossType 'L1'