Research map / Image generation / Image restoration
Dual-Domain Self-Supervised Learning and Model Adaption for Deep Compressive Imaging
Yuhui Quan, Xinran Qin, Tongyao Pang, Hui Ji
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add image-domain loss L_Image (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 34.11 -> 36.37 | Improved |
| add measurement-domain loss L_Measure (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 16.67 -> 36.37 | Improved |
| add outer noise injection (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 28.20 -> 36.37 | Improved |
| use sign-flipping noise instead of Gaussian noise for injected r (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 35.67 -> 36.37 | Improved |
| replace self-supervised loss with supervised l2 loss (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 36.37 -> 36.98 | No clear effect |
| add test-time model adaption (ablation) | Ours-NA · Sparse-view CT (CT100) | PSNR (dB): 41.35 -> 41.99 | Improved |
| add test-time model adaption (ablation) | Ours-NA · MRI radial mask, CS ratio 1/4, noise 0.1*MaxValue | PSNR (dB): 30.62 -> 31.04 | Improved |
| add test-time model adaption (ablation) | Ours-NA · Set11 NIR (noise 0, CS ratio 0.25) | PSNR (dB): 32.70 -> 33.41 | Improved |
| add test-time model adaption when trained on radial masks and tested on Gaussian 25% (ablation) | Ours-NA · MRI cross-mask | PSNR (dB): 35.18 -> 35.86 | Improved |
| swap in ISTANet+ architecture with test-time adaption (ablation) | Ours-TA · Set11 NIR (noise 0, CS ratio 0.4) | PSNR (dB): 36.67 -> 37.18 | Mixed |
| add test-time adaption instead of per-sample internal learning (BCNN) | Ours-TA · BSD68 running time (CS 40%) | running time (s): 90013 -> 2178 | Improved |
| add dual-domain loss versus LDAMP-SURE | Ours-NA · Set11 NIR (noise 0, CS ratio 0.25) | PSNR (dB): 31.37 -> 32.70 | Improved |
Closest papers by meaning
- Self-supervised Bayesian Deep Learning for Image Recovery with Applications to Compressive Sensing (ECCV 2020 · similarity 0.89)
- Self-Supervised Neural Networks for Spectral Snapshot Compressive Imaging (ICCV 2021 · similarity 0.89)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.86)
- UNSURE: self-supervised learning with Unknown Noise level and Stein's Unbiased Risk Estimate (ICLR 2025 · similarity 0.86)
- Using Powerful Prior Knowledge of Diffusion Model in Deep Unfolding Networks for Image Compressive Sensing (CVPR 2025 · similarity 0.86)
- CPP-Net: Embracing Multi-Scale Feature Fusion into Deep Unfolding CP-PPA Network for Compressive Sensing (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
