Research map / 3D vision / Face and shape
Metadata-Based RAW Reconstruction via Implicit Neural Functions
Leyi Li, Huijie Qiao, Qi Ye, Qinmin Yang
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 |
|---|---|---|---|
| Replace CAM with INF using uniform RAW sampling | RAW reconstruction · CAM dataset [24], three cameras | PSNR: 49.57 -> 53.53; 51.54 -> 56.71; 53.11 -> 58.36 | Improved |
| Replace CAM with INF on full-resolution images | RAW reconstruction · NUS [9], three cameras | PSNR: 49.51 -> 61.48; 52.87 -> 63.38; 53.34 -> 63.80 | Improved |
| Use INF instead of RBF interpolation (ablation) | RAW reconstruction with sRGB input · RAW reconstruction | PSNR: 29.10 -> 59.62 | Improved |
| Use INF instead of RBF interpolation (ablation) | RAW reconstruction with coordinate input · RAW reconstruction | PSNR: 21.71 -> 31.09 | Improved |
| Use INF instead of RBF interpolation with combined inputs (ablation) | RAW reconstruction · RAW reconstruction | PSNR: 49.69 -> 60.91 | Improved |
| Separate sRGB values and coordinates into distinct branches rather than combine them (ablation) | INF · RAW reconstruction | PSNR: 60.91 -> 62.24 | Improved |
| Add sRGB values as input alongside coordinates (ablation) | INF · RAW reconstruction | PSNR: 31.09 -> 60.91 | Improved |
| Use patch-specific reconstruction instead of larger patches (ablation) | INF · RAW reconstruction, regularization λ=(0.0001, 0.1, 0.001) | PSNR: 59.64 -> 62.24 | Improved |
| Use patch size ×1 instead of ×2 (ablation) | INF · RAW reconstruction, regularization λ=(0.0001, 0.1, 0.001) | PSNR: 62.24 -> 62.58 | Improved |
| Add regularization with λ=(0.0001, 0.1, 0.001) at patch size ×2 (ablation) | INF · RAW reconstruction | PSNR: 49.17 -> 62.24 | Improved |
| Use ReLU activation instead of sine activation (ablation) | INF · RAW reconstruction | PSNR | Hurt |
| Use ReLU activation instead of sine activation (ablation) | INF · Guided depth super-resolution on Middlebury | RMSE: 2.50 -> 2.39; 3.06 -> 2.69; 5.23 -> 3.57 | Improved |
Closest papers by meaning
- Learning sRGB-to-Raw-RGB De-Rendering With Content-Aware Metadata (CVPR 2022 · similarity 0.92)
- Beyond Learned Metadata-Based Raw Image Reconstruction (IJCV 2024 · similarity 0.91)
- RAW Image Reconstruction Using a Self-Contained sRGB-JPEG Image With Only 64 KB Overhead (CVPR 2016 · similarity 0.90)
- Raw Image Reconstruction With Learned Compact Metadata (CVPR 2023 · similarity 0.90)
- Leveraging Frame Affinity for sRGB-to-RAW Video De-rendering (CVPR 2024 · similarity 0.87)
- ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge (CVPR 2025 · 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.
