MIDV-DM: A Document-Oriented Dataset for Image Manipulation Detection and Localization

MIDV-DM dataset is described in a paper “MIDV-DM: A Document-Oriented Dataset for Image Manipulation Detection and Localization” published in 2025 by researchers from Smart Engines (smartengines.com) and Shenzhen MSU-BIT University (en.smbu.edu.cn).

MIDV-DM is a publicly available dataset aimed at detecting and localizing manipulations in identity document images. It contains 1000 original document images from the previously published MIDV-2020 dataset and 8000 automatically created manipulated images covering eight types of forgeries. These manipulations were conceptually categorized based on the analysis of over 2000 real-world fraud attempts. The dataset facilitates development, testing, and benchmarking of methods for document image authenticity analysis. 

The corresponding paper describes eight manipulation types and explains the process of obtaining baseline quality based on the IML-ViT model.