Research map / Video / Video generation
Conditional Image-to-Video Generation With Latent Flow Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use latent-flow diffusion instead of direct video synthesis | cI2V generation · MUG, MHAD, NATOPS | FVD: MUG: 108.02 -> 27.57; MHAD: 280.26 -> 152.48; NATOPS: 169.61 -> 160.84 | Helped |
| use latent-flow diffusion instead of direct video synthesis | cI2V generation at 128×128 · MUG, MHAD, NATOPS | FVD: MUG: 126.28 -> 32.09; MHAD: 337.43 -> 214.39; NATOPS: 344.81 -> 195.17 | Helped |
| finetune decoder Ω on unlabeled new-domain videos (ablation) | LFDM · FaceForensics | FID: 38.48 -> 23.36 | Helped |
| finetune decoder Ω on unlabeled new-domain videos (ablation) | LFDM · FaceForensics | sFID: 75.08±37.33 -> 51.56±26.36 | Helped |
| finetune decoder Ω on unlabeled new-domain videos (ablation) | LFDM · FaceForensics | L1 error: 2.129 -> 1.310 | Helped |
| use DDIM-10 instead of DDPM-1000 (ablation) | LFDM_128 · MUG | FVD: 32.09 -> 50.18 | Hurt |
| use DDIM-10 instead of DDPM-1000 (ablation) | LFDM_128 · MUG | sampling time: 36.0 -> 0.3 | Helped |
| use DDIM-100 instead of DDPM-1000 (ablation) | LFDM_128 · MUG | FVD: 32.09 -> 106.11 | Hurt |
| increase guidance scale from 1.0 to 1.1 (ablation) | LFDM_128 · MUG | FVD: 32.09 -> 31.84 | Helped |
| increase guidance scale from 1.0 to 1.1 (ablation) | LFDM_128 · MUG | sampling time: 36.0 -> 71.4 | Hurt |
| increase guidance scale from 1.0 to 5.0 (ablation) | LFDM_128 · MUG | FVD: 32.09 -> 49.82 | Hurt |
Closest papers by meaning
- Decouple Content and Motion for Conditional Image-to-Video Generation (AAAI 2024 · similarity 0.91)
- Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model (NeurIPS 2024 · similarity 0.90)
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.89)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- TRIP: Temporal Residual Learning with Image Noise Prior for Image-to-Video Diffusion Models (CVPR 2024 · similarity 0.88)
- MeDM: Mediating Image Diffusion Models for Video-to-Video Translation with Temporal Correspondence Guidance (AAAI 2024 · similarity 0.88)
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.