Research map / Image generation / Image restoration
Basis Prediction Networks for Effective Burst Denoising With Large Kernels
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use burst-specific low-dimensional spatiotemporal basis kernels and per-pixel coefficients | BPN · Synthetic grayscale test set, Gain ∝ 1 | Average PSNR (dB): 36.88 -> 38.18 | Helped |
| Use burst-specific low-dimensional spatiotemporal basis kernels and per-pixel coefficients | BPN · Synthetic color test set, Gain ∝ 1 | Average PSNR (dB): 38.86 -> 40.16 | Helped |
| Use kernel size K=15 instead of K=5 (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 35.70 -> 36.29 | Helped |
| Use kernel size K=15 instead of K=9 (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.22 -> 36.29 | Helped |
| Increase basis size from 10 to 90 kernels (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 35.31 -> 36.29 | Helped |
| Increase basis size from 50 to 90 kernels (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.10 -> 36.29 | Helped |
| Increase basis size from 90 to 130 kernels (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.29 -> 36.27 | Hurt |
| Replace the spatiotemporal basis with a common spatial basis across frames (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.29 -> 35.71 | Hurt |
| Replace the spatiotemporal basis with separate spatial bases per frame (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.29 -> 36.21 | Hurt |
| Use a fixed, input-agnostic basis instead of predicting a burst-specific basis (ablation) | BPN · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 36.29 -> 34.66 | Hurt |
| Use a low-dimensional basis decomposition instead of unconstrained per-pixel KPN kernels (ablation) | Burst denoising · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 34.29 -> 36.29 | Helped |
| Use a low-dimensional subspace rather than separable spatial kernels (ablation) | Burst denoising · Validation dataset, Gain ∝ 1 | Average PSNR (dB): 34.67 -> 36.29 | Helped |
Closest papers by meaning
- Burst Denoising With Kernel Prediction Networks (CVPR 2018 · similarity 0.90)
- gp2Scale: A Class of Compactly Supported Non-Stationary Kernels and Distributed Computing for Exact Gaussian Processes on 10 Million Data Points (ICML 2026 · similarity 0.85)
- Next-Scale Prediction: A Self-Supervised Approach for Real-World Image Denoising (CVPR 2026 · similarity 0.85)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.83)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.83)
- ManiFusion: Unlocking High-Throughput Generation via Superposition in Manifold Space (NeurIPS 2026 · similarity 0.83)
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.