Research map / Video / Video segmentation and tracking
SAFNet: Selective Alignment Fusion Network for Efficient HDR Imaging
Lingtong Kong, Bo Li, Yike Xiong, Hao Zhang, Hong Gu, Jinwei Chen
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 selection masks M_i to the first fusion stage alongside optical flow (ablation) | SAFNet first stage · Kalantari 17 | PSNR-mu: 33.69 -> 41.68 | Improved |
| add optical flow to the first fusion stage alongside selection masks (ablation) | SAFNet first stage · Kalantari 17 | PSNR-mu: 40.69 -> 41.68 | Improved |
| feed merged HDR image H_m into the refinement network (ablation) | SAFNet refine stage · Kalantari 17 | PSNR-mu: 43.87 -> 44.59 | Improved |
| feed optical flow and selection masks into the refinement network (ablation) | SAFNet refine stage · Kalantari 17 | PSNR-mu: 43.63 -> 43.87 | Improved |
| use window partition cropping with 512 input for stage 1 and 128 for stage 2 (ablation) | SAFNet · Kalantari 17 | PSNR-mu: 44.59 -> 44.66 | Improved |
| increase training patch size from 256 to 512 | SAFNet vs NHDRRNet · Challenge123 | accuracy | Mixed |
| coarse-to-fine fully convolutional design without attention | SAFNet · Kalantari 17 at 1500x1000 | running time: about 10x, 18x, 23x faster than FlexHDR, HDR-Transformer, SCTNet | Improved |
| SAFNet vs transformer methods on Tel 23 (near-static backgrounds) | SAFNet · Tel 23 test | PSNR-mu: 44.93 -> 44.61 (SCTNet vs SAFNet) | Hurt |
| SAFNet vs SCTNet on Tel 23 in linear domain | SAFNet · Tel 23 test | PSNR-l: 51.73 -> 51.97 | Improved |
Later papers that compared with this method
1 later accepted paper reports SAFNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DeAltHDR: Learning HDR Video Reconstruction from Degraded Alternating Exposure Sequences (ICLR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Unified HDR Imaging Method With Pixel and Patch Level (CVPR 2023 · similarity 0.89)
- Exposure-Aware Dynamic Weighted Learning for Single-Shot HDR Imaging (ECCV 2022 · similarity 0.88)
- HDR Video Reconstruction: A Coarse-To-Fine Network and a Real-World Benchmark Dataset (ICCV 2021 · similarity 0.88)
- Towards Real-World HDR Video Reconstruction: A Large-Scale Benchmark Dataset and A Two-Stage Alignment Network (CVPR 2024 · similarity 0.87)
- HDRFlow: Real-Time HDR Video Reconstruction with Large Motions (CVPR 2024 · similarity 0.87)
- Self-Supervised High Dynamic Range Imaging with Multi-Exposure Images in Dynamic Scenes (ICLR 2024 · similarity 0.86)
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.
