Model weights, code, and reproduction artifacts for Team MSU's third-place
solution to the ACM MM 2026 GenText-Forensics challenge
(Codabench).
The method is a decomposed chain-of-thought pipeline for detecting,
localizing, typing, and explaining forgeries in multilingual document text
images:
DTD (Document Tampering Detector) — an external pixel-level visual
tampering detector that produces a tampering… See the full description on the dataset page:
https://huggingface.co/datasets/cmcshnik/GenText-Forensics_third_place_additional_materials.