Efficient transformer with compressed-attention for stereo image super-resolution

Dependencies

• Python 3.9
• PyTorch 1.10.0

cd code
pip install -r requirements.txt
python setup.py develop

Datasets

• LCATSSR

Training Set	Testing Set
Flickr1024 + Middlebury	KITTI2012 + KITTI2015 + Middlebury + Flickr1024

• LCATSR

Training Set	Testing Set
DIV2K	Set5 + Set14 + BSD100 + Urban100 + Manga109

Refer to the datasets folder for the complete data.

Implementation of LCATSSR

Train

#scale factor 2
python -m torch.distributed.launch --nproc_per_node=4 --master_port=4321 basicsr/train.py -opt options/train/LCATSSR/LCATSSR_x2.yml --launcher pytorch
#scale factor 4
python -m torch.distributed.launch --nproc_per_node=4 --master_port=4321 basicsr/train.py -opt options/train/LCATSSR/LCATSSR_x4.yml --launcher pytorch

Test

#scale factor 2
python -m torch.distributed.launch --nproc_per_node=1 --master_port=4321 basicsr/test.py -opt options/test/LCATSSR/LCATSSR_x2.yml --launcher pytorch
#scale factor 4
python -m torch.distributed.launch --nproc_per_node=1 --master_port=4321 basicsr/test.py -opt options/test/LCATSSR/LCATSSR_x4.yml --launcher pytorch

Implementation of LCATSR

Train

#scale factor 2
python -m torch.distributed.launch --nproc_per_node=2 --master_port=4321 basicsr/train.py -opt options/train/LCATSR/LCATSR_x2.yml --launcher pytorch
#scale factor 3
python -m torch.distributed.launch --nproc_per_node=2 --master_port=4321 basicsr/train.py -opt options/train/LCATSR/LCATSR_x3.yml --launcher pytorch
#scale factor 4
python -m torch.distributed.launch --nproc_per_node=2 --master_port=4321 basicsr/train.py -opt options/train/LCATSR/LCATSR_x4.yml --launcher pytorch

Test

#scale factor 2
python test_SISR.py --scale 2 --model_path './experiments/pretrained_models/LCATSR_x2.pth'
#scale factor 3
python test_SISR.py --scale 3 --model_path './experiments/pretrained_models/LCATSR_x3.pth'    
#scale factor 4
python test_SISR.py --scale 4 --model_path './experiments/pretrained_models/LCATSR_x4.pth'  

Citation

@article{song2025etcassr,
  title = {Efficient transformer with compressed-attention for stereo image super-resolution},
  author={Song, Jianwen and Sowmya, Arcot and Zhang, Weichuan and Sun, Changming},
  journal = {Knowledge-Based Systems},
  year = {2025}