Summary
AEGIS is a holistic benchmark introduced for evaluating forensic analysis of AI-generated academic images. It offers three key advances over existing benchmarks. First, it covers domain-specific complexity across seven academic categories and 39 fine-grained subtypes, revealing intrinsic forensic difficulty: even GPT-5.1 achieves only 48.80% overall performance, while expert models attain limited localization accuracy (IoU 30.09%). Second, it simulates diverse forgeries by modeling four prevalent academic forgery strategies across 25 generative models, 11 of which yield average forensic accuracy below 50%, indicating that forensic techniques lag behind generative advances. Third, it enables multidimensional forensic evaluation jointly assessing detection, inference, and localization, exposing complementary strengths across model families: multimodal large language models (MLLMs) reach 84.74% accuracy on identifying textual artifacts, while expert detectors peak at 79.54% on binary authenticity detection. Evaluating 25 leading MLLMs, 9 expert models, and 1 unified multimodal understanding-and-generation model, AEGIS serves as a diagnostic testbed revealing fundamental limitations in academic image forensics. Paper: arXiv 2604.28177.
Paper Overview
Field: Computer Vision
Authors: Shilin Lu, Qinying Huang, Kai Wang et al.
Posted: 2026-04-30
arXiv: 2604.28177
Abstract
We introduce AEGIS, a holistic benchmark for Evaluating forensic analysis of AI-Generated academic ImageS. Compared to existing benchmarks, AEGIS features three key advances:
1. Domain-Specific Complexity: covering seven academic categories with 39 fine-grained subtypes, exposing intrinsic forensic difficulty, where even GPT-5.1 reaches 48.80% overall performance and expert models achieve only limited localization accuracy (IoU 30.09%).
2. Diverse Forgery Simulations: modeling four prevalent academic forgery strategies across 25 generative models, 11 of which yield average forensic accuracy below 50%, indicating that forensic techniques lag behind generative advances.
3. Multidimensional Forensic Evaluation: jointly assessing detection, inference, and localization, revealing complementary strengths across model families — MLLMs reach 84.74% accuracy on textual artifact identification, while expert detectors peak at 79.54% on binary authenticity detection.
Through evaluating 25 leading MLLMs, 9 expert models, and 1 unified multimodal understanding-and-generation model, AEGIS serves as a diagnostic testbed exposing fundamental limitations in academic image forensics.
---
*Auto-collected on 2026-05-02*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177619036