HDiff-HIR: Hierarchically Conditional Diffusion Model for Hyperspectral Image Reconstruction

This repo is the implementation of HDiff-HIR: Hierarchically Conditional Diffusion Model for Hyperspectral Image Reconstruction (TCSVT2025).

Comparison on Simulation Dataset

The performance are reported on 10 scenes of the KAIST dataset. The test size of FLOPS is 256 x 256.

Quantitative Results

Method	Params (M)	FLOPS (G)	PSNR	SSIM
TSA-Net	44.25	110.06	31.46	0.894
HDNet	2.37	154.76	34.97	0.943
MST	2.03	28.15	35.18	0.948
CST	3.00	40.10	36.12	0.957
HDiff-HIR (ours)	13.80	74.48	36.88	0.966

Qualitative Results

Download results of HDiff-HIR (Google Drive).

Usage

Prepare Dataset

Download cave_1024_28 (Baidu Disk, code: fo0q | One Drive), CAVE_512_28 (Baidu Disk, code: ixoe | One Drive), KAIST_CVPR2021 (Baidu Disk, code: 5mmn | One Drive), TSA_simu_data (Baidu Disk, code: efu8 | One Drive), TSA_real_data (Baidu Disk, code: eaqe | One Drive), and then put them into the corresponding folders of datasets/ and recollect them as the following form:

|--HDiff-HIR
    |--real
        |-- dataset.py
        |-- option.py
        |-- train.py
        |-- sample.py
    |--simulation
        |-- dataset.py
        |-- option.py
        |-- train.py
        |-- sample.py
|--datasets
    |--cave_1024_28
        |--scene1.mat
        ：  
    |--CAVE_512_28
        |--scene1.mat
        ：  
    |--KAIST_CVPR2021  
        |--1.mat
        ： 
    |--TSA_simu_data  
        |--mask.mat   
        |--Truth
            |--scene01.mat
            ： 
    |--TSA_real_data  
        |--mask.mat   
        |--Measurements
            |--scene1.mat
            ： 

Following CST and DAUHST, we use the CAVE dataset (cave_1024_28) as the simulation training set. Both the CAVE (CAVE_512_28) and KAIST (KAIST_CVPR2021) datasets are used as the real training set.

Simulation Experiement

Training

cd HDiff-HIR/simulation/
python train.py --method hdiff-l

The training log, trained model, and reconstructed HSI will be saved in HDiff-HIR/simulation/exp/test/ (configurable via --outf in option.py).

Testing (Sampling)

cd HDiff-HIR/simulation/
python sample.py --method hdiff-l

The sampled results will be saved to the directory specified in the sample.py script.

Real Experiement

Training

cd HDiff-HIR/real/
python train.py --method hdiff-l

The training log, trained model, and reconstructed HSI will be saved in HDiff-HIR/real/real/diffusion/.

Testing (Sampling)

cd HDiff-HIR/real/
python sample.py --method hdiff-l

The sampled results will be output into the corresponding test directory inside real/.

Visualization

For the visualization toolkit, please see DSMT-main.

Citation

If this code helps you, please consider citing our work:

@ARTICLE{luo2026hdiff,
  author={Luo, Fulin and Chen, Xi and Fu, Chuan and Guo, Tan and Du, Bo},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={HDiff-HIR: Hierarchically Conditional Diffusion Model for Hyperspectral Image Reconstruction}, 
  year={2026},
  volume={36},
  number={1},
  pages={777-791},
}