Research map / Video / Video generation
Swap the Right Identity: Spatio-Temporal Preference Optimization for Identity Swapping
Hongdeng Shen, Xiongzheng Li, Jikang Cheng, Duo Li, Yunlong FENG, Jing Li
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 DPO preference optimization instead of continual SFT on chosen samples (ablation) | Continual SFT · Character swapping benchmark | ID-Sim: 0.4356 -> 0.4959 | Improved |
| add Diffusion DPO on top of the SFT base model (ablation) | SFT base model · Character swapping benchmark | PoseErr: 144.47 -> 107.59 | Improved |
| add Region-Weighted Preference (RWP) to DPO (ablation) | Diffusion DPO · Character swapping benchmark | ID-Sim: 0.4959 -> 0.5168 | Improved |
| add RWP to DPO (ablation) | Diffusion DPO · Character swapping benchmark | PoseErr: 107.59 -> 99.10 | Improved |
| add RWP to DPO (ablation) | Diffusion DPO · Character swapping benchmark | BG-CLIP: 0.7470 -> 0.7416 | Hurt |
| add RWP to DPO (ablation) | Diffusion DPO · Character swapping benchmark | FG-CLIP: 0.8689 -> 0.8792 | Improved |
| add Time-Gated Identity Preference (TIP) to DPO+RWP (ablation) | DPO + RWP · Character swapping benchmark | ID-Sim: 0.5168 -> 0.5975 | Improved |
| add TIP to DPO+RWP (ablation) | DPO + RWP · Character swapping benchmark | BG-CLIP: 0.7416 -> 0.7328 | Hurt |
| add TIP to DPO+RWP (ablation) | DPO + RWP · Character swapping benchmark | FID: 50.65 -> 51.80 | Hurt |
| add TIP to DPO+RWP (ablation) | DPO + RWP · Character swapping benchmark | BG-SSIM: 0.9199 -> 0.9277 | Improved |
| use SIPA-tuned Qwen-Image-Edit instead of original Qwen-Image-Edit 2509 | Qwen-Image-Edit 2509 · Character swapping benchmark | PoseErr: 220.1150 -> 100.0920 | Improved |
| use SIPA instead of GPT-Image 1.5 for identity | GPT-Image 1.5 · Character swapping benchmark | ID-Sim: 0.6133 -> 0.5975 | Hurt |
Closest papers by meaning
- DynamicFace: High-Quality and Consistent Face Swapping for Image and Video using Composable 3D Facial Priors (ICCV 2025 · similarity 0.89)
- High-Fidelity Diffusion Face Swapping with ID-Constrained Facial Conditioning (CVPR 2026 · similarity 0.89)
- CanonSwap: High-Fidelity and Consistent Video Face Swapping via Canonical Space Modulation (ICCV 2025 · similarity 0.88)
- Optimizing ID Consistency in Multimodal Large Models: Facial Restoration via Alignment, Entanglement, and Disentanglement (ICLR 2026 · similarity 0.88)
- Reinforced Disentanglement for Face Swapping without Skip Connection (ICCV 2023 · similarity 0.87)
- DreamSalon: A Staged Diffusion Framework for Preserving Identity-Context in Editable Face Generation (CVPR 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.
