Research map / Image generation / Image restoration
Self-supervised Image Denoising with Downsampled Invariance Loss and Conditional Blind-Spot Network
Yeong Il Jang, Keuntek Lee, Gu Yong Park, Seyun Kim, Nam Ik Cho
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 the conditional blind-spot network so center pixel is visible at inference (ablation) | C-BSN · SIDD validation | PSNR (dB): 35.86 -> 36.22 | Improved |
| use random subsampler downsampling instead of Noise2Same-style masking only (ablation) | C-BSN · SIDD validation | PSNR (dB): 25.58 -> 36.22 | Improved |
| use L1 norm instead of RMS in the invariance loss (ablation) | C-BSN · SIDD validation | PSNR (dB): 35.63 -> 36.22 | Improved |
| use random subsampler instead of pixel-shuffle downsampling at stride 2 (ablation) | C-BSN · SIDD validation | PSNR (dB): 35.32 -> 36.22 | Improved |
| use random subsampler instead of space2batch at stride 2 (ablation) | C-BSN · SIDD validation | PSNR (dB): 36.02 -> 36.22 | Improved |
| increase RS stride from 2 to 5 in invariance loss (ablation) | C-BSN · SIDD validation | PSNR (dB): 36.22 -> 35.24 | Hurt |
| add blind self-supervised loss with warm-up (ablation) | C-BSN · SIDD validation | PSNR (dB): 25.92 -> 36.22 | Improved |
| add warm-up scheduling for the C-BSN loss weight (ablation) | C-BSN · SIDD validation | PSNR (dB): 35.65 -> 36.22 | Improved |
| use space2batch stride 5 for blind loss instead of stride 2 (ablation) | C-BSN · SIDD validation | PSNR (dB): 25.93 -> 36.22 | Improved |
| train on test set noisy images instead of SIDD medium | C-BSN · SIDD benchmark | PSNR (dB): 36.82 -> 36.84 | No clear effect |
| replace AP-BSN with C-BSN without post-processing | AP-BSN · SIDD benchmark | PSNR (dB): 34.90 -> 36.82 | Improved |
| change RS stride from 3/4 combination to 2/5 baseline | C-BSN · SIDD validation | PSNR (dB) | No clear effect |
Closest papers by meaning
- Exploring Efficient Asymmetric Blind-Spots for Self-Supervised Denoising in Real-World Scenarios (CVPR 2024 · similarity 0.94)
- TM-BSN: Triangular-Masked Blind-Spot Network for Real-World Self-Supervised Image Denoising (CVPR 2026 · similarity 0.94)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.93)
- Rethinking Transformer-Based Blind-Spot Network for Self-Supervised Image Denoising (AAAI 2025 · similarity 0.92)
- Pseudo-Siamese Blind-spot Transformers for Self-Supervised Real-World Denoising (NeurIPS 2024 · similarity 0.91)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.90)
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.
