Research map / Image generation / Image restoration
Dual-Output Multi-Exposure HDR Reconstruction via SDR Fusion and Gain Map Inverse Tone Mapping
Jinho Kim, Jinwoo Kim, Seon Joo Kim
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 |
|---|---|---|---|
| jointly train SDR synthesis and gain map instead of cascading pretrained UltraFusion and GMNet (ablation) | DOME-HDR · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 22.96 -> 44.63 | Improved |
| jointly train SDR synthesis and gain map instead of cascading pretrained UltraFusion and GMNet (ablation) | DOME-HDR · HDR reconstruction (Challenge123+Tel+Kalantari) | LPIPS: 0.1042 -> 0.0156 | Improved |
| use separate cross-attention branches per exposure instead of normalized summation (ablation) | UltraFusion · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 43.76 -> 44.63 | Improved |
| use separate cross-attention branches (ablation) | UltraFusion · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-SSIM: 0.9876 -> 0.9878 | No clear effect |
| use separate cross-attention branches (ablation) | UltraFusion · HDR reconstruction (Challenge123+Tel+Kalantari) | LPIPS: 0.0180 -> 0.0156 | Improved |
| add under- and over-exposed images to HDR prior construction (SDR only to all three) (ablation) | HPGM · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 23.69 -> 44.63 | Improved |
| add UE to SDR+ME inputs for HDR prior (ablation) | HPGM · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 23.69 -> 25.32 | Improved |
| use OE, ME instead of SDR only for HDR prior (ablation) | HPGM · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 23.69 -> 26.16 | Improved |
| add all three exposures instead of two (ablation) | HPGM · HDR reconstruction (Challenge123+Tel+Kalantari) | LPIPS: 0.1041 -> 0.0156 | Improved |
| add Phase 2 training on MEF-generated SDR after Phase 1 (ablation) | DOME-HDR · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 40.53 -> 44.63 | Improved |
| add Phase 1 joint training before Phase 2 (ablation) | DOME-HDR · HDR reconstruction (Challenge123+Tel+Kalantari) | PU-PSNR: 43.26 -> 44.63 | Improved |
| add Phase 2 to Phase 1 (ablation) | DOME-HDR · HDR reconstruction (Challenge123+Tel+Kalantari) | CLIPIQA: 0.4180 -> 0.4137 | Hurt |
Closest papers by meaning
- GMODiff: One-Step Gain Map Refinement with Diffusion Priors for Efficient HDR Reconstruction (ECCV 2026 · similarity 0.91)
- AFUNet: Cross-Iterative Alignment-Fusion Synergy for HDR Reconstruction via Deep Unfolding Paradigm (ICCV 2025 · similarity 0.89)
- Single Shot HDR Recovery via a Video Diffusion Prior (NeurIPS 2026 · similarity 0.89)
- Zero-Shot Structure-Preserving Diffusion Model for High Dynamic Range Tone Mapping (CVPR 2024 · similarity 0.89)
- Generating Content for HDR Deghosting from Frequency View (CVPR 2024 · similarity 0.88)
- GaussHDR: High Dynamic Range Gaussian Splatting via Learning Unified 3D and 2D Local Tone Mapping (CVPR 2025 · similarity 0.87)
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.
