sahebi/basic-super-resolution

Basic Super Resolution

This git cloned from https://github.com/icpm/super-resolution change some modifications.

• Add CARN method
• Add different optimization method
• Log the checkppoints and _logs
• Log the result

Optimizer

• ADAM
• AdamSparse
• Adamax
• Adadelta
• Adagrad
• ASGD
• LAMB
• RProp
• SGD
• RMSprop

Single Image Super Resolution Methods

• SubPixelCNN
• SRCNN
• SRCNNT
• VDSR
• EDSR
• FSRCNN
• DRCN batchsize should be small, batchsize=4
• SRGAN change save checkpoint path patterns
• DBPN batchsize should be small, batchsize=1
• MemNet
• CARN

Mount google drive install basic super resolution package

import os
from google.colab import drive
drive.mount('/content/gdrive/')

!pip install tensorboardX

os.chdir('/content/gdrive/My Drive/Projects/master-code')
!git clone https://github.com/sahebi/basic-super-resolution

Run train all python script

os.chdir('/content/gdrive/My Drive/Projects/master-code/basic-super-resolution')

!python train_all.py --logprefix test_BSDS300_2x -uf 2 --dataset BSDS300 --batchSize 128 --testBatchSize 128 --nEpochs 1500 --iter 3

Train Model

python train.py --logprefix testmodel -uf 4 --dataset BSDS300 --batchSize 16 --testBatchSize 8 --nEpochs 1 --model srcnnt

python train.py --logprefix test1epoch -uf 4 --dataset COCO --batchSize 16 --testBatchSize 8 --nEpochs 1 --model srcnnt

Run super resolution

python super_resolve.py --input result/BSD300_3096.jpg/4x/lr.jpg --model model/carn_488.pth --output result/BSD300_3096.jpg/4x/carn.jpg