Awesome

GMSR: Gradient-Guided Mamba for Spectral Reconstruction from RGB Images

[Xinying Wang], [Zhixiong Huang], [Sifan Zhang], [Jiawen Zhu], [Paolo Gamba], and [Lin Feng]

GMSR-Net Framework

Spectral Reconstruction

We propose GMSR-Net, the first Mamba framework designed for SR task. We are going to enlarge our model zoo in the future. The support list is as follows:

• RepCPSI (TGRS 2023)
• MST++ (CVPRW 2022)
• MST++ (CVPRW 2022)
• AGD-Net (TCI 2021)
• HSRNet (TNNLS 2020)
• HRNet (CVPRW 2020)
• MSCNN (PRCV 2018)

Train

1. Created Environment.

• anaconda NVIDIA GPU
• python-3.10.13
• torch-2.1.1+cu118
• pip install causal_conv1d==1.1.1
• pip install mamba_ssm==1.2.0

2. Data Preprocess.

Before training, you need to split the original datasets into 128*128 by train_data_preprocess and valid_data_preprocess. Then, replace the data location in the file. Finally, you can obtain the running data with:

Getting the prepared train data by run:

python train_data_preprocess.py --data_path './data/Dataset' --patch_size 128 --stride 64 --train_data_path './dataset/2020Train'

Getting the prepared valid data by run:

python valid_data_preprocess.py --data_path './data/Dataset' --valid_data_path './dataset/2020Val'

Training.

python main.py

The data generated during training will be recorded in /results/GMSR.

Test

python test.py

• Download the checkpoints ( Baidu Disk, code: GMSR))

Citation

If you find this code helpful, please kindly cite:

# GMSR-Net
@article{wang2024gmsr,
  title={GMSR: Gradient-Guided Mamba for Spectral Reconstruction from RGB Images},
  author={Wang, Xinying and Huang, Zhixiong and Zhang, Sifan and Zhu, Jiawen and Feng, Lin},
  journal={arXiv preprint arXiv:2405.07777},
  year={2024}
}