Lightweight Image Super-Resolution Method Based on Mamba-Transformer Fusion Architecture

Visual Results On MTSR

Receptive Field

Inference Speed

Installation

The training experiment are conducted using PyTorch with two NVIDIA 4090 GPUs.

1. Configure Your Environment (Prerequisites)

• Ubuntu 22.04
• Python 3.8
• PyTorch 2.0.1
• CUDA 11.7+
• causal_conv1d 1.0.0
• opencv-python
• mamba_ssm 1.0.1

Note: It may also work with other versions.

2. Datasets

• Training Dataset:
  • DIV2K
• Testing Datasets:
  • Set5, Set14, BSD100, Urban100
  • Manga109

Create Environment

conda env create -f MTSR.yml
conda activate MTSR

Pretrained Models

Pretrained models are available in the Releases.

Training

• Cropped input size: 64×64
• GPUs: 2
• Batch size: 16 per GPU

python -m torch.distributed.launch --nproc_per_node=2 --master_port=1234 basicsr/train.py -opt options/train/train_MTSR_x2.yml --launcher pytorch
python -m torch.distributed.launch --nproc_per_node=2 --master_port=1234 basicsr/train.py -opt options/train/train_MTSR_x3.yml --launcher pytorch
python -m torch.distributed.launch --nproc_per_node=2 --master_port=1234 basicsr/train.py -opt options/train/train_MTSR_x4.yml --launcher pytorch

Testing

python basicsr/test.py -opt options/test/test_MTSR_x2.yml
python basicsr/test.py -opt options/test/test_MTSR_x3.yml
python basicsr/test.py -opt options/test/test_MTSR_x4.yml

License

This project is released under the Apache 2.0 license.

Acknowledgement

This project is based on BasicSR and MambaIR.
Thanks to the authors for their outstanding contributions.