Research map / Image generation / Text-to-image diffusion
Early Timestep Zero-Shot Candidate Selection for Instruction-Guided Image Editing
Joowon Kim, Ziseok Lee, Donghyeon Cho, Sanghyun Jo, Yeonsung Jung, Kyungsu Kim, Eunho Yang
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 |
|---|---|---|---|
| select seed by BIS from multiple seeds | IP2P · PIE-Bench | background MSE x10^4: 248.493 -> 127.481 | Improved |
| select seed by BIS from multiple seeds | UltraEdit · PIE-Bench | background MSE x10^4: 87.544 -> 63.847 | Improved |
| early stop at t_stop=60 instead of full denoising for best-of-N (ablation) | IP2P · PIE-Bench | background MSE x10^4: 146.151 -> 127.481 | Improved |
| early stop at t_stop=60 instead of full denoising for best-of-N (ablation) | MagicBrush · PIE-Bench | PSNR: 26.253 -> 26.133 | Mixed |
| early stop at t_stop=60 instead of full denoising for best-of-N (ablation) | UltraEdit · PIE-Bench | CLIP-T: 25.379 -> 25.362 | Mixed |
| use early-timestep ELECT instead of Best of N (N=8) (ablation) | five editing models · PIE-Bench | NFE: 41% average reduction, up to 61% | Improved |
| add MLLM prompt selection after seed selection | IP2P · PIE-Bench | failure ratio: 40.00% -> 28.57% | Improved |
| add MLLM prompt selection after seed selection | MagicBrush · PIE-Bench | failure ratio: 26.71% -> 16.57% | Improved |
| add MLLM prompt selection after seed selection | UltraEdit · PIE-Bench | failure ratio: 23.43% -> 17.00% | Improved |
| use continuous squared mean relevance mask instead of thresholded mask (ablation) | ELECT · PIE-Bench | seed selection robustness | Improved |
| use CLIP or highest FG MSE as early-step selection cue instead of BG MSE (ablation) | ELECT · PIE-Bench | selection quality | Hurt |
| increase seeds N | ELECT · PIE-Bench | background MSE | Mixed |
Closest papers by meaning
- PnP Inversion: Boosting Diffusion-based Editing with 3 Lines of Code (ICLR 2024 · similarity 0.87)
- From Scale to Speed: Adaptive Test-Time Scaling for Image Editing (CVPR 2026 · similarity 0.87)
- NEP: Autoregressive Image Editing via Next Editing Token Prediction (NeurIPS 2025 · similarity 0.87)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.87)
- Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer (NeurIPS 2025 · similarity 0.87)
- TiNO-Edit: Timestep and Noise Optimization for Robust Diffusion-Based Image Editing (CVPR 2024 · 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.
