ABS

This repository contains the source code and data for the paper "Adaptive Branch Selection for Accelerate Image Super-Resolution" in The Visual Computer.

Citation

If you have used this code or data in your research, please cite the following papers:

@article{
  title   = {Adaptive Branch Selection for Accelerate Image Super-Resolution}
  author  = {Cheng, Ding and Zhongqiu, Zhao and Hao, Shen and Xiufeng Liu}
  year    = {2025}}

How to Train ABS

Environmental dependencies

python 3.7 + PyTorch 1.13 + CUDA11.6

Data Preparation

Train

DIV2K: https://data.vision.ee.ethz.ch/cvl/DIV2K/

get Sub-images

Replace the dataset path in the ..data2patch/data_scripts/extract_subimages.py file, and then run the codes as follows:

cd ../data2patch/data_scripts/
python extract_subimages.py

Test

Test2K,Test4K,Test8K:https://drive.google.com/drive/folders/18b3QKaDJdrd9y0KwtrWU2Vp9nHxvfTZH

First Stage

The First Stage is to train the three branches. Train the basic SR model as Hard-Branch, and run the codes as follows:

cd ../code/
python basicsr/train.py -opt options/train/train_base.yml

To train the SR branch with fewer channels, replace the teacher model path in the ..options/train/train_base_d.yml, and run the codes as follows:

python basicsr/train.py -opt options/train/train_base_d.yml

Second-Stage

The Second-Stage is to train the regressor, replace the three branch path in the ..options/train/train_ABS.yml and run the codes as follows:

python basicsr/train.py -opt options/train/train_ABS.yml