Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised Scalable Deep Compressed Sensing
Bin Chen, Xuanyu Zhang, Shuai Liu, Yongbing Zhang, Jian Zhang
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 stage-2 cross-image internal learning (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 32.36 -> 33.09 (+0.73) | Improved |
| add stage-3 single-image internal learning after stage 2 (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 33.09 -> 33.56 (+0.47) | Improved |
| add stage-4 self-ensemble after stage 3 (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 33.56 -> 33.72 (+0.16) | Improved |
| add stage-1 external pretraining before stage 3 (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 31.49 -> 33.56 | Improved |
| add supervised DOC loss for ratio scalability (ablation) | supervised l2 SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 18.62 -> 32.76 (+14.14) | Improved |
| use l1 (p=1) instead of l2 (p=2) (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 31.26 -> 32.40 | Improved |
| add symmetric loss terms (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 32.40 -> 32.36 | No clear effect |
| add geometric transformations T (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 32.34 -> 32.40 | No clear effect |
| randomize the CS ratio in DOC loss instead of fixing it (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 13.02 -> 32.34 | Improved |
| randomize division ratios instead of fixing both at 5% (ablation) | SC-CNN · Set11 (trained at ratio 10%, avg over test ratios) | PSNR (dB): 10.33 -> 18.34 | Improved |
| add random measurement division (DMC) to MC loss (ablation) | SC-CNN · Toy signals (ratio 50%, avg of 4 sets) | SNR (dB): 11.77 -> 14.65 | Improved |
| use DOC instead of OC in MC+OC (ablation) | SC-CNN · Toy signals (ratio 50%, avg of 4 sets) | SNR (dB): 16.93 -> 28.74 | Improved |
Closest papers by meaning
- Scalable Convolutional Neural Network for Image Compressed Sensing (CVPR 2019 · similarity 0.90)
- Practical Compact Deep Compressed Sensing (TPAMI 2024 · similarity 0.89)
- Deep Compressed Sensing (ICML 2019 · similarity 0.88)
- A Data-Driven and Distributed Approach to Sparse Signal Representation and Recovery (ICLR 2019 · similarity 0.87)
- Learning a Compressed Sensing Measurement Matrix via Gradient Unrolling (ICML 2019 · similarity 0.87)
- Multi-Cross Sampling and Frequency-Division Reconstruction for Image Compressed Sensing (AAAI 2024 · similarity 0.85)
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.
