LocateEdit-Bench: A Benchmark for Instruction-Based Editing Localization

2026-02-05

AI Summary

This work addresses the challenge of detecting and localizing forgeries introduced by emerging instruction-driven image editing, which existing forensic methods—primarily designed for traditional inpainting-based manipulations—struggle to handle effectively. To bridge this gap, the authors introduce LocateEdit-Bench, the first benchmark specifically tailored for forgery localization in instruction-guided editing scenarios. It comprises 231,000 images generated by four state-of-the-art editing models across three common editing types. The study also establishes a comprehensive multi-metric evaluation protocol to systematically assess current localization approaches. By providing a high-quality, open-source dataset and a rigorous evaluation framework, this work fills a critical void in the forensic analysis of instruction-based image manipulations and highlights the limitations of existing techniques, thereby laying a foundation for future research in this domain.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity
• Responsible Web: Measurement, analysis, and circumvention of Web censorship

Abstract

Recent advancements in image editing have enabled highly controllable and semantically-aware alteration of visual content, posing unprecedented challenges to manipulation localization. However, existing AI-generated forgery localization methods primarily focus on inpainting-based manipulations, making them ineffective against the latest instruction-based editing paradigms. To bridge this critical gap, we propose LocateEdit-Bench, a large-scale dataset comprising $231$K edited images, designed specifically to benchmark localization methods against instruction-driven image editing. Our dataset incorporates four cutting-edge editing models and covers three common edit types. We conduct a detailed analysis of the dataset and develop two multi-metric evaluation protocols to assess existing localization methods. Our work establishes a foundation to keep pace with the evolving landscape of image editing, thereby facilitating the development of effective methods for future forgery localization. Dataset will be open-sourced upon acceptance.

Problem

• instruction-based editing
• manipulation localization
• image forgery
• editing localization
• AI-generated images

Innovation

• instruction-based editing
• forgery localization
• benchmark dataset
• image manipulation
• editing localization