Research map / Video / Video generation
LiveMoments: Reselected Key Photo Restoration in Live Photos via Reference-guided Diffusion
Clara Xue, Zizheng Yan, Zhenning Shi, Yuhang Yu, Jingyu Zhuang, Qi Zhang, Jinwei Chen, Qingnan Fan
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 ReferenceNet (reference guidance from original key photo) (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9081 -> 0.9539 | Improved |
| add ReferenceNet (ablation) | LiveMoments · SynLive260 | LPIPS: 0.1685 -> 0.0837 | Improved |
| add motion-guided attention instead of no latent alignment (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9539 -> 0.9629 | Improved |
| use motion-guided attention instead of warping reference image (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9480 -> 0.9629 | Improved |
| use motion-guided attention instead of warping reference latent (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9437 -> 0.9629 | Improved |
| warp reference image instead of motion-guided attention (ablation) | LiveMoments · vivoLive144 | NIQE_re: 0.0990 (full) vs 0.1034 (warp image) vs 0.1097 (no alignment) | Mixed |
| add motion alignment on synthetic data (ablation) | LiveMoments · SynLive260 | PSNR: 31.84 (warp latent) -> 31.65 (full) | Hurt |
| train with customized Live Photo degradation instead of StableSR degradation (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9581 -> 0.9629 | Improved |
| train with customized degradation instead of SeeSR degradation (ablation) | LiveMoments · vivoLive144 | NIQE_re: 0.0904 -> 0.0990 | Hurt |
| inject 200% Gaussian noise into optical flow (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9629 -> 0.9497 | Hurt |
| replace RAFT with SPyNet (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9629 -> 0.9629 | No clear effect |
| replace RAFT with LiteFlowNet (ablation) | LiveMoments · vivoLive144 | DINO-Q: 0.9629 -> 0.9595 | No clear effect |
Closest papers by meaning
- Perceiving Better Moments: Cover Frame Reselection and Enhancement for Live Photos with the Live2K Dataset (ECCV 2026 · similarity 0.89)
- Harnessing Image Diffusion Prior for Photo-Realistic Video Restoration (NeurIPS 2026 · similarity 0.86)
- ReFIR: Grounding Large Restoration Models with Retrieval Augmentation (NeurIPS 2024 · similarity 0.85)
- Self-Supervised Selective-Guided Diffusion Model for Old-Photo Face Restoration (NeurIPS 2025 · similarity 0.85)
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.85)
- RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution (AAAI 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.
