ML Security PapersA Comprehensive Dataset for Human vs. AI Generated Image Detection
Rajarshi Roy 1, Nasrin Imanpour 2, Ashhar Aziz 3, Shashwat Bajpai 4, Gurpreet Singh 5, Shwetangshu Biswas 6, Kapil Wanaskar 7, Parth Patwa 8, Subhankar Ghosh 9, Shreyas Dixit 10, Nilesh Ranjan Pal 1, Vipula Rawte 2, Ritvik Garimella 2, Gaytri Jena 11, Vasu Sharma 12, Vinija Jain 12, Aman Chadha 13, Aishwarya Naresh Reganti 13, Amitava Das 14
Published on arXiv
2601.00553
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Releases a 96,000-sample benchmark spanning five modern generative models, enabling standardized evaluation of AI-generated image detectors and model-attribution classifiers.
MS COCOAI
Novel technique introduced
Multimodal generative AI systems like Stable Diffusion, DALL-E, and MidJourney have fundamentally changed how synthetic images are created. These tools drive innovation but also enable the spread of misleading content, false information, and manipulated media. As generated images become harder to distinguish from photographs, detecting them has become an urgent priority. To combat this challenge, We release MS COCOAI, a novel dataset for AI generated image detection consisting of 96000 real and synthetic datapoints, built using the MS COCO dataset. To generate synthetic images, we use five generators: Stable Diffusion 3, Stable Diffusion 2.1, SDXL, DALL-E 3, and MidJourney v6. Based on the dataset, we propose two tasks: (1) classifying images as real or generated, and (2) identifying which model produced a given synthetic image. The dataset is available at https://huggingface.co/datasets/Rajarshi-Roy-research/Defactify_Image_Dataset.
Key Contributions
- MS COCOAI dataset of 96,000 real and synthetic images generated by five state-of-the-art models (SD 3, SD 2.1, SDXL, DALL-E 3, MidJourney v6)
- Two benchmark tasks: binary real-vs-AI classification and multi-class model attribution (identifying which generator produced an image)
- Baseline experiments establishing initial performance benchmarks on both tasks
🛡️ Threat Analysis
The paper directly supports AI-generated content detection (deepfake/synthetic image detection), which is explicitly within ML09 scope. The dataset and tasks are designed to evaluate detectors for synthetic images produced by DALL-E 3, Stable Diffusion, MidJourney, etc.