Research map / Image generation / Text-to-image diffusion
Collaborative Score Distillation for Consistent Visual Editing
Subin Kim, Kyungmin Lee, June Suk Choi, Jongheon Jeong, Kihyuk Sohn, Jinwoo Shin
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 SVGD kernel mixing across samples (ablation) | CSD-Edit · Video editing ablation | CLIP image consistency: 0.909 -> 0.957 | Improved |
| add SVGD kernel mixing across samples (ablation) | CSD-Edit · Video editing ablation | LPIPS: 0.372 -> 0.236 | Improved |
| subtract image-conditioned noise instead of random noise (ablation) | CSD-Edit · Video editing ablation | CLIP directional similarity: 0.270 -> 0.320 | Improved |
| subtract image-conditioned noise instead of random noise (ablation) | CSD-Edit · Video editing ablation | CLIP image consistency: 0.884 -> 0.957 | Improved |
| add SVGD kernel mixing across samples (ablation) | CSD-Edit · 3D scene editing ablation | CLIP directional similarity: 0.186 -> 0.239 | Improved |
| subtract image-conditioned noise instead of random noise (ablation) | CSD-Edit · 3D scene editing ablation | CLIP directional similarity: 0.186 -> 0.239 | Improved |
| replace FateZero by CSD-Edit | video editing · Video editing (DAVIS) | LPIPS: 0.264 -> 0.235 | Improved |
| replace IN2N by CSD-Edit for 3D scene editing | NeRF editing · 3D scene editing | CLIP directional similarity: 0.177 -> 0.215 | Improved |
| replace SDS by CSD in DreamFusion | text-to-3D · Text-to-3D (DreamFusion) | FID: 259.4 -> 247.1 | Improved |
| increase minibatch size B from 4 to 12 (ablation) | CSD-Edit panorama · Panorama editing | CLIP image similarity: 0.240 -> 0.195 | Hurt |
| increase minibatch size B from 4 to 12 (ablation) | CSD-Edit panorama · Panorama editing | iterations per second: 2.86 -> 1.02 | Hurt |
| use fewer iterations with higher learning rate (ablation) | CSD-Edit video · Video editing (DAVIS) | CLIP image consistency: 0.955 vs 0.957 | No clear effect |
Closest papers by meaning
- Rethinking Score Distillation as a Bridge Between Image Distributions (NeurIPS 2024 · similarity 0.89)
- Invertible Consistency Distillation for Text-Guided Image Editing in Around 7 Steps (NeurIPS 2024 · similarity 0.88)
- Posterior Distillation Sampling (CVPR 2024 · similarity 0.88)
- Taming Mode Collapse in Score Distillation for Text-to-3D Generation (CVPR 2024 · similarity 0.88)
- Rethinking Score Distilling Sampling for 3D Editing and Generation (ICML 2025 · similarity 0.87)
- Consistent Flow Distillation for Text-to-3D Generation (ICLR 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.
