Single Image Super Resolution Model Zoo

Project Introduction

This is a project for image super resolution. In this project, we implemented some state-of-art methods for image super resolution. It contains the dataset preprocessing, model training, model evaluating. Every subdir contains the codes for training and testing of one specific method.

Setup

Prerequisites

  • Tensorflow 1.1.1
  • Linux with Nvidia GPU + cuDNN

Geting Started

git clone git@github.com:BingzheWu/tf-sr-zoo.git
cd tf-sr-zoo

Then, chose the model you want to use, and get in the subdir to see more details.

Some Results

News

Date Update
20170720 Collect all trace of Evaluation
20170715 Add attention Layer
20170710 Add Instance Normalization
20170708 Add evaluation module
20170705 Add gan based pix2pix model
20170703 Add LapSRN model
20170628 TFRecord Creator
20170624 Add Dataset Module

To-Do List

  • Dataset Module (Done)
  • Lap-SRN Method (Done)
  • Pix2Pix Method (Done)
  • Perceptual-GAN (Done)
  • Instance Normalization(Done)
  • Add unify testing module for every datasets (Doing)
  • Add vis Tools (To Do)
  • Combine L2 + L1 + adversarial loss with weighted term(To Do)
  • Build a Docker Image (To Do)