Research map / Learning theory and algorithms / Kernels and regression theory
Deep Physics-Guided Unrolling Generalization for Compressed Sensing
Bin Chen, Jiechong Song, Jingfen Xie, Jian Zhang
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 feature-domain unrolling (ablation) | OPINE-Net+ · Set11 10% | PSNR: 29.81 -> 30.10 | Improved |
| add feature-domain unrolling (ablation) | MADUN · Set11 10% | PSNR: 29.89 -> 30.16 | Improved |
| add encoder-to-decoder and intra-stage skip connections (ablation) | PRL-PGD · Set11 10% | PSNR: 31.04 -> 31.49 | Improved |
| add encoder-to-decoder and intra-stage skip connections (ablation) | PRL-RND · Set11 10% | PSNR: 18.57 -> 31.46 | Improved |
| replace fixed gradient with learned 3x3 Conv + Sigmoid fusion (ablation) | PRL-PGD · Urban100 30% | PSNR: 33.98 -> 34.21 | Improved |
| add physics-engaged channels (q from 0 to 1) (ablation) | PRL-PGD+ D=16 · Set11 10% | PSNR: 30.67 -> 31.54 | Improved |
| increase physical feature dimensionality q beyond 1 (ablation) | PRL-PGD+ D=16 · Set11 10% | PSNR: 31.54 -> 31.68 | No clear effect |
| use full-channel physics fusion with q=64 in plain FD network (ablation) | plain FD PGD network · Set11 10% | PSNR: 0.74dB | Improved |
| use FD-improved high-throughput stage in plain FD architecture (ablation) | plain FD network · Set11 | PSNR: 0.88dB | Improved |
| use adaptive gradient descent step sizes (ablation) | PRL networks · Set11 | PSNR | No clear effect |
| use multiscale U-shaped PRL instead of plain architecture (ablation) | Plain(FD)-PGD · Set11 50% | PSNR and inference time: 41.53 (201.092 ms) -> 41.78 (72.067 ms) | Improved |
| use multiscale PRL instead of single-scale PRL* (ablation) | PRL*-PGD · Set11 50% | inference time: 548.295 -> 72.067 | Improved |
Closest papers by meaning
- Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging (CVPR 2025 · similarity 0.88)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.87)
- UFC-Net: Unrolling Fixed-point Continuous Network for Deep Compressive Sensing (CVPR 2024 · similarity 0.87)
- CPP-Net: Embracing Multi-Scale Feature Fusion into Deep Unfolding CP-PPA Network for Compressive Sensing (CVPR 2024 · similarity 0.87)
- Unrolled denoising networks provably learn to perform optimal Bayesian inference (NeurIPS 2024 · similarity 0.87)
- Using Powerful Prior Knowledge of Diffusion Model in Deep Unfolding Networks for Image Compressive Sensing (CVPR 2025 · similarity 0.87)
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.
