Awesome

[CVPR2024] HIR-Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion Models

Li Pang, Xiangyu Rui, Long Cui, Hongzhong Wang, Deyu Meng, Xiangyong Cao

Abstract: Hyperspectral image (HSI) restoration aims at recovering clean images from degraded observations and plays a vital role in downstream tasks. Existing model-based methods have limitations in accurately modeling the complex image characteristics with handcraft priors, and deep learning-based methods suffer from poor generalization ability. To alleviate these issues, this paper proposes an unsupervised HSI restoration framework with pre-trained diffusion model (HIR-Diff), which restores the clean HSIs from the product of two low-rank components, i.e., the reduced image and the coefficient matrix. Specifically, the reduced image, which has a low spectral dimension, lies in the image field and can be inferred from our improved diffusion model where a new guidance function with total variation (TV) prior is designed to ensure that the reduced image can be well sampled. The coefficient matrix can be effectively pre-estimated based on singular value decomposition (SVD) and rank-revealing QR (RRQR) factorization. Furthermore, a novel exponential noise schedule is proposed to accelerate the restoration process (about 5x acceleration for denoising) with little performance decrease. Extensive experimental results validate the superiority of our method in both performance and speed on a variety of HSI restoration tasks, including HSI denoising, noisy HSI super-resolution, and noisy HSI inpainting.

Hightlights

Environment

conda create -n hirdiff python=3.9
conda activate hirdiff
pip3 install -r requirements.txt

HSI Restoration

Download the pretrained diffusion model

downloading the pretrained diffusion model I190000_E97_gen.pth provided by ddpm-cd and put the model into checkpoints\diffusion

Download the data

downloading the data for denoise, super-resolution and inpainting from Google Drive or Baidu Netdisk(code:fzst).

Testing

Denoising on Houston dataset (sigma=50)

python main.py -eta1 16 -eta2 10 --k 8 -step 20 -dn Houston --task denoise --task_params 50 -gpu 0 --beta_schedule exp  

Super-Resolution on WDC dataset (upscale factor=x4)

python main.py -eta1 500 -eta2 12 --k 8 -step 20 -dn WDC --task sr --task_params 0.25 -gpu 0 --beta_schedule exp 

Inpainting on Salinas dataset (masking rate=0.8)

python main.py -eta1 8 -eta2 6 --k 5 -step 20 -dn Salinas --task inpainting --task_params 0.8 -gpu 0 --beta_schedule exp 

The effectiveness of RRQR

To disable RRQR, adding '--no_rrqr' and the bands are selected at equal intervals

python main.py -eta1 16 -eta2 10 --k 8 -step 20 -dn Houston --task denoise --task_params 50 -gpu 0 --no_rrqr
python main.py -eta1 500 -eta2 12 --k 8 -step 20 -dn WDC --task sr --task_params 0.25 -gpu 0 --no_rrqr
python main.py -eta1 8 -eta2 6 --k 5 -step 20 -dn Salinas --task inpainting --task_params 0.8 -gpu 0 --no_rrqr

Comparison with other schedules

# linear schedule (psnr: 34.33)
python main.py -eta1 20 -eta2 1 -step 20 -dn Houston --task denoise --task_params 50 -gpu 0 --beta_schedule linear      

# cosine schedule (psnr: 34.61)
python main.py -eta1 20 -eta2 1 -step 20 -dn Houston --task denoise --task_params 50 -gpu 0 --beta_schedule cosine      

# exponential schedule (psnr: 36.01)
python main.py -eta1 16 -eta2 10 --k 8 -step 20 -dn Houston --task denoise --task_params 50 -gpu 0 --beta_schedule exp   

Citation

@article{pang2024hir,
  title={HIR-Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion Models},
  author={Pang, Li and Rui, Xiangyu and Cui, Long and Wang, Hongzhong and Meng, Deyu and Cao, Xiangyong},
  journal={arXiv preprint arXiv:2402.15865},
  year={2024}
}

Contact