Reconstructed Convolution Module Based Look-Up Tables for Efficient Image Super-Resolution

[Guandu Liu*], Yukang Ding, Mading Li, Ming Sun, Xing Wen and [Bin Wang#]

Efficiency

Overview

The core idea of our paper is RC Module.

Usage

Our code follows the architecture of MuLUT. In the sr directory, we provide the code of training RC-LUT networks, transferring RC-LUT network into LUts, finetuning LUTs, and testing LUTs, taking the task of single image super-resolution as an example. In the common/network.py, RC_Module is the core module of our paper.

Dataset

Please following the instructions of training. And you can also prepare SRBenchmark

Installation

Clone this repo

git clone https://github.com/liuguandu/RC-LUT

Install requirements: torch>=1.5.0, opencv-python, scipy

Train

First, please train RC network follow next code

sh ./sr/5x57x79x9MLP_combined.sh

updating...