Research map / Image generation / Editing and style transfer
DLEBench: Evaluating Small-scale Object Editing Ability for Instruction-based Image Editing Model
Shibo Hong, Boxian Ai, Jun Kuang, Wei Wang, FengJiao Chen, Zhongyuan Peng, Chenhao Huang, Yixin Cao
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 |
|---|---|---|---|
| use Oracle-guided Mode instead of LMM-as-a-Judge (ablation) | Gemini-3-Pro evaluator · DLEBench human alignment | correlation with human judgments / MAE | Improved |
| use Tool-driven Mode (Grounding, Zoom-In, Difference, Enhancer tools) instead of plain LMM-as-a-Judge (ablation) | Gemini-3-Pro evaluator · DLEBench human alignment | correlation with human judgments / MAE | Improved |
| swap the judge backbone from Gemini-3-Pro to GPT-4.1 in plain LMM-as-a-Judge (ablation) | LMM-as-a-Judge · DLEBench human alignment | correlation with human judgments | No clear effect |
| use crop-and-edit with adaptive bbox expansion for reference images | reference image generation · DLEBench construction | reference quality | Improved |
| use closed-source model over open-source | IIEMs · DLEBench | average score: Gemini-3-Pro 65.55; Bagel-Think 61.00 vs GPT-Image-1 40.25 | Mixed |
| edit multiple small objects (Change Count) instead of one | all IIEMs · DLEBench | average score: Gemini-3-Pro 43.87 vs 65.55 average | Hurt |
| larger target object area | most IIEMs · DLEBench | Instruction Following score | Improved |
| larger target object area | UniWorld-V1, Qwen-Edit, MagicBrush · DLEBench | Instruction Following score | No clear effect |
| conservative strategy of abstaining from editing under uncertainty | Bagel-Think vs GPT-Image-1 · DLEBench | Visual Consistency score: 35.17 -> 86.43 | Improved |
Closest papers by meaning
- I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing (NeurIPS 2024 · similarity 0.92)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.89)
- 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)
- SmartEdit: Exploring Complex Instruction-based Image Editing with Multimodal Large Language Models (CVPR 2024 · similarity 0.88)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (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.
