GitXplorerGitXplorer
p

super_resolution

public
8 stars
1 forks
0 issues

Commits

List of commits on branch master.
Verified
89908a7e99c0ba03fc8d6f2c52cecccc4eed8a86

edited readme

pprajjwal1 committed 5 years ago
Unverified
928bbdcaf73276e39a5f702f6ee9749e5075ac4a

u

pprajjwal1 committed 6 years ago
Unverified
a10272be2e0d466f8d79b70c18fab89e98709c1d

update

pprajjwal1 committed 6 years ago
Unverified
14e8afba6185953ec5c05c55e15b7404661f33f1

update

pprajjwal1 committed 6 years ago
Unverified
6322fc602c8cfcc6195a8ea8af2c9e4c77b26e57

fixed name

pprajjwal1 committed 6 years ago
Unverified
2001f68094223dfeb7def1d476765ae4759e51aa

add *

pprajjwal1 committed 6 years ago

README

The README file for this repository.

This repo contains implementations of papers

Super Resolution

Two methods are being used to perform super resolution

  • EDSR + Pixel Shuffling + Perceptual Loss
  • Pre-trained GAN (traning Generator and critic seperately and then training both)

Requirements

  • Pytorch>=0.3
  • Fastai

Usage

$ python super.py

Dataset

  • ImageNet 10% sample
  • Celeb A

EDSR

Requires Fastai==0.7 (super_res_edsr.py)

In this method, We're making use of Pixel shuffle along with EDSR (Removal of BN in skip connection)

Original image

Prediction 

Input image

Prediction 

.

Pre Trained GANs

PreRequisites

  • FastAI>=1.45
  • Pytorch >= 1.0

Creating Low quality dataset

Use crappification from (utils.py)

We use a Unet Generator which has Resnet34/50 (Pretrained) as it's head (modifiable) Generator is trained with low quality images with ground truth being images from original dataset.

When pretraining generator only, here's the output:

Input----------------------------------------------------Prediction------------------------------------------Target

gen_pre1 gen_pre2 gen_pre3

Critic is trained on generator images and original images

Usage

Check the path variables carefully

$ python3 super_res_gan.py 

References