DETER: Detecting Edited Regions for Deterring Generative Manipulations

Sai Wang, Ye Zhu, Ruoyu Wang, Amaya Dharmasiri, Olga Russakovsky, Yu Wu

Keywords: Deepfake detection, regional manipulation, dataset and benchmark

TL;DR: We introduce a large-scale image dataset constructed with state-of-the-art generators for regional manipulations in different granularities.

Abstract: Generative AI capabilities have grown substantially in recent years, raising renewed concerns about the potential malicious use of generated data, or "deep fakes." Despite being a longstanding and important research topic, deep fake detection research on most existing datasets has not kept pace with generative AI advancements sufficiently to develop detection technology that can meaningfully alert human users in real-world settings. In this work, we introduce DETER, a large-scale dataset for DETEcting edited image Regions and deterring modern advanced generative manipulations. After a comprehensive study of prior literature, our proposed dataset makes contributions along three main axes: the upgrade on modern manipulations via the state-of-the-art generative models; the mitigation of biased spurious correlations in prior deep fake datasets; and a more unified formulation suitable for various detection models in different granularities. Equipped with DETER, we conduct extensive experiments and detailed analysis using our rich annotations and improved benchmark protocols, revealing future directions and the next set of challenges in developing reliable regional fake detection models.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2400