This repository provides a comprehensive toolbox for spectral compressive imaging reconstruction, offering implementations for over 15 algorithms, including MST, CST, DAUHST, and BiSRNet. It serves researchers and practitioners in hyperspectral imaging, enabling them to reconstruct spectral data from compressed measurements using advanced deep learning and model-based techniques.

How It Works

The toolbox leverages transformer architectures and unfolding techniques to address spectral compressive imaging challenges. Methods like MST utilize mask-guided spectral-wise transformers for efficient reconstruction, while DAUHST incorporates degradation-aware unfolding and half-shuffle mechanisms. This approach aims to achieve state-of-the-art performance in terms of reconstruction accuracy (PSNR, SSIM) and computational efficiency.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: NVIDIA GPU with CUDA, Python 3 (Anaconda recommended).
• Dataset: Requires downloading and organizing datasets like cave_1024_28, CAVE_512_28, KAIST_CVPR2021, etc., into a specific directory structure.
• Pretrained Models: Downloadable from provided Google Drive/Baidu Disk links.
• Documentation: Detailed instructions for simulation and real-world experiments are available in the README.

Highlighted Details

• Supports 16 algorithms including 12 learning-based and 3 model-based methods.
• MST++ won the NTIRE 2022 Challenge on spectral recovery from RGB images.
• Provides quantitative comparisons (PSNR, SSIM) and model complexity metrics (Params, FLOPS).
• Includes code for training, testing, and visualization for both simulation and real-world data.

Maintenance & Community

The project is actively maintained with frequent updates and releases, including new algorithms and participation in recent challenges (e.g., NTIRE 2024). Links to papers and arXiv preprints are provided for each method.

Licensing & Compatibility

The repository does not explicitly state a license. However, the inclusion of research papers and the nature of the code suggest it is intended for research purposes. Commercial use would require clarification on licensing terms.

Limitations & Caveats

The README does not specify a license, which may pose a barrier to commercial adoption. Some dataset downloads are via Baidu Disk, which might have regional access limitations. The setup requires careful organization of multiple datasets.