ForensicZip: More Tokens are Better but Not Necessary in Forensic Vision-Language Models

2026-03-12

AI Summary

This work addresses the high computational cost of existing vision-language models in high-resolution multimedia forensics and the tendency of conventional token pruning methods to discard forgery traces due to their semantic bias. The authors propose a training-free token compression framework that, for the first time, approaches token selection from the perspective of forgery detection. By integrating a Birth-Death optimal transport model with relaxed dummy nodes and a high-frequency spectral prior, the method dynamically scores and retains tokens capturing physical discontinuities indicative of manipulation. With only 10% of tokens preserved, the approach achieves a 2.97× inference speedup and over 90% reduction in FLOPs, while maintaining state-of-the-art performance in both deepfake and AIGC-generated content detection.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Cyber-crime defenses and forensics

Abstract

Multimodal Large Language Models (MLLMs) enable interpretable multimedia forensics by generating textual rationales for forgery detection. However, processing dense visual sequences incurs high computational costs, particularly for high-resolution images and videos. Visual token pruning is a practical acceleration strategy, yet existing methods are largely semantic-driven, retaining salient objects while discarding background regions where manipulation traces such as high-frequency anomalies and temporal jitters often reside. To address this issue, we introduce ForensicZip, a training-free framework that reformulates token compression from a forgery-driven perspective. ForensicZip models temporal token evolution as a Birth-Death Optimal Transport problem with a slack dummy node, quantifying physical discontinuities indicating transient generative artifacts. The forensic scoring further integrates transport-based novelty with high-frequency priors to separate forensic evidence from semantic content under large-ratio compression. Experiments on deepfake and AIGC benchmarks show that at 10\% token retention, ForensicZip achieves $2.97\times$ speedup and over 90\% FLOPs reduction while maintaining state-of-the-art detection performance.

Problem

• multimedia forensics
• visual token pruning
• forgery detection
• high-frequency anomalies
• temporal jitters

Innovation

• ForensicZip
• visual token pruning
• optimal transport
• multimodal forensics
• training-free compression