Research map / Image generation / Editing and style transfer
Semantically Aligned Gradient-Driven Context-Preserving Image Editing
Chiranjeev Chiranjeev, Muskan Dosi, MAYANK VATSA, RICHA SINGH
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 distillation loss L_D (gradient-enabled VLM supervision) to denoising loss (ablation) | IABEdit · RealEdit | harmonic mean (CLIP-T, CS-P): +2.47 HM | Improved |
| add LoRA adaptation to the Instruction Aligner (ablation) | IABEdit · RealEdit | harmonic mean (CLIP-T, CS-P): +4.42 HM | Improved |
| replace LLaVA-7B with Qwen-VL-2.5-7B (ablation) | IABEdit · RealEdit | harmonic mean (CLIP-T, CS-P): 42.00 -> 42.27 | Improved |
| add context-preservation conditioning (ablation) | IABEdit · RealEdit | harmonic mean (CLIP-T, CS-P): +1.81 HM | Improved |
| raise timestep threshold tau from 200 to 300 (ablation) | IABEdit · RealEdit | harmonic mean (CLIP-T, CS-P): 42.00 -> 40.67 | Hurt |
| add distillation loss L_D on D-LORD (ablation) | IABEdit · D-LORD (face editing) | harmonic mean (CLIP-T, CS-P): 41.71 -> 44.68 | Improved |
| add identity coherence loss L_C (ablation) | IABEdit · D-LORD (face editing) | DINO-P: 53.59 -> 55.33 | Improved |
| train FLUX.1 with IABEdit | FLUX.1 Kontext Dev · RealEdit | harmonic mean (CLIP-T, CS-P): 41.29 -> 42.28 | Improved |
| IABEdit vs best diffusion baseline UltraEdit | IABEdit · MagicBrush | DINO-I: 84.77 -> 88.26 | Improved |
| IABEdit vs SuperEdit on GPT-4o overall rating | IABEdit · RealEdit (GPT-4o) | overall score: 3.91 -> 3.78 | Hurt |
| IABEdit vs Gemini-AI agent on identity preservation | IABEdit · D-LORD (face editing) | DINO-P: 50.20 -> 55.33 | Improved |
| IABEdit vs SuperEdit on CS-P | IABEdit · RealEdit | CS-P: 90.93 -> 89.65 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SuperEdit — SuperEdit: Rectifying and Facilitating Supervision for Instruction-Based Image Editing (ICCV 2025) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Improving Editability in Image Generation with Layer-wise Memory (CVPR 2025 · similarity 0.90)
- CAMILA: Context-Aware Masking for Image Editing with Language Alignment (NeurIPS 2025 · similarity 0.90)
- FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · similarity 0.90)
- Balancing Preservation and Modification: A Region and Semantic Aware Metric for Instruction-Based Image Editing (ICML 2025 · similarity 0.89)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.89)
- Instruct-CLIP: Improving Instruction-Guided Image Editing with Automated Data Refinement Using Contrastive Learning (CVPR 2025 · 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.
