Incode - DeepID workshop

Renat Khizbullin, Lazar Lazarevic, Alexey Tochin, Aleksei Grishin, Anastasiia Evteeva

Keywords: deep learning, forgery detection

Abstract: Document forgery detection is increasingly critical due to the rise of deepfake and image manipulation technologies. We present Incode's solution for the ICCV 2025 DeepID challenge, which combines three complementary detectors: a general-purpose forgery detector (TruFor), a portrait manipulation detection model, and a text alteration detection model. Our system uses domain-adaptive thresholding for robust decision-making across synthetic and real-world identity documents, achieving second place overall in the challenge. Extensive ablation studies highlight the contribution of each component and the importance of domain calibration.

Submission Number: 2