Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable Deepfake Detection with RL Enhanced Self-Blended Images

Ning Jiang, Dingheng Zeng, Yanhong Liu, Haiyang Yi, Shijie Yu, Minghe Weng, Haifeng Shen, Ying Li

TL;DR

The paper tackles the lack of explainability in deepfake detection by integrating multimodal language models with reinforcement learning to produce interpretable CoT outputs. It introduces Self-Blended Images (SBI) to automatically generate forged data and a text-based forgery localization reward within a GRPO framework to mitigate reward sparsity, enabling effective RL for a binary detection task. A two-stage training pipeline—Supervised Fine-Tuning on CoT-rich data followed by GRPO-based reinforcement learning—delivers competitive cross-dataset performance without dedicated detectors, while producing precise forgery clues. The approach reduces annotation costs, strengthens cross-domain generalization, and provides explainable insights into detected forgeries, with implementation details available publicly.

Abstract

Most prior deepfake detection methods lack explainable outputs. With the growing interest in multimodal large language models (MLLMs), researchers have started exploring their use in interpretable deepfake detection. However, a major obstacle in applying MLLMs to this task is the scarcity of high-quality datasets with detailed forgery attribution annotations, as textual annotation is both costly and challenging - particularly for high-fidelity forged images or videos. Moreover, multiple studies have shown that reinforcement learning (RL) can substantially enhance performance in visual tasks, especially in improving cross-domain generalization. To facilitate the adoption of mainstream MLLM frameworks in deepfake detection with reduced annotation cost, and to investigate the potential of RL in this context, we propose an automated Chain-of-Thought (CoT) data generation framework based on Self-Blended Images, along with an RL-enhanced deepfake detection framework. Extensive experiments validate the effectiveness of our CoT data construction pipeline, tailored reward mechanism, and feedback-driven synthetic data generation approach. Our method achieves performance competitive with state-of-the-art (SOTA) approaches across multiple cross-dataset benchmarks. Implementation details are available at https://github.com/deon1219/rlsbi.

Explainable Deepfake Detection with RL Enhanced Self-Blended Images

TL;DR

The paper tackles the lack of explainability in deepfake detection by integrating multimodal language models with reinforcement learning to produce interpretable CoT outputs. It introduces Self-Blended Images (SBI) to automatically generate forged data and a text-based forgery localization reward within a GRPO framework to mitigate reward sparsity, enabling effective RL for a binary detection task. A two-stage training pipeline—Supervised Fine-Tuning on CoT-rich data followed by GRPO-based reinforcement learning—delivers competitive cross-dataset performance without dedicated detectors, while producing precise forgery clues. The approach reduces annotation costs, strengthens cross-domain generalization, and provides explainable insights into detected forgeries, with implementation details available publicly.

Abstract

Most prior deepfake detection methods lack explainable outputs. With the growing interest in multimodal large language models (MLLMs), researchers have started exploring their use in interpretable deepfake detection. However, a major obstacle in applying MLLMs to this task is the scarcity of high-quality datasets with detailed forgery attribution annotations, as textual annotation is both costly and challenging - particularly for high-fidelity forged images or videos. Moreover, multiple studies have shown that reinforcement learning (RL) can substantially enhance performance in visual tasks, especially in improving cross-domain generalization. To facilitate the adoption of mainstream MLLM frameworks in deepfake detection with reduced annotation cost, and to investigate the potential of RL in this context, we propose an automated Chain-of-Thought (CoT) data generation framework based on Self-Blended Images, along with an RL-enhanced deepfake detection framework. Extensive experiments validate the effectiveness of our CoT data construction pipeline, tailored reward mechanism, and feedback-driven synthetic data generation approach. Our method achieves performance competitive with state-of-the-art (SOTA) approaches across multiple cross-dataset benchmarks. Implementation details are available at https://github.com/deon1219/rlsbi.
Paper Structure (11 sections, 2 equations, 2 figures, 3 tables)

This paper contains 11 sections, 2 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Construction of Basic Forged Data
  5. Construction of CoT Data
  6. Text-Based Forgery Localization Reward
  7. Model Tuning
  8. Supervised Fine-Tuning
  9. Reinforcement Training
  10. Experiment
  11. Conclusion

Figures (2)

  • Figure 1: The pipeline of generating CoT data. First, using a randomly selected mask and perturbation parameters to generate a fake image. Then retrieve key captions from the lookup table based on the mask and parameters. Finally, feed the image pair and key captions into an MLLM to produce the corresponding CoT annotation.
  • Figure 2: The feedback-based GRPO training. Dynamically adjust the data generation parameters based on computed historical reward values.