AB
AB
AiBoss

All Categories

project

LOKI - A synthetic data detection benchmark jointly launched by Sun Yat-sen University and Shanghai AI Lab

LOKI is a synthetic data detection benchmark jointly proposed by Sun Yat-sen University and Shanghai AI Lab. It aims to comprehensively evaluate the ability of large-scale multimodal models (LMMs) to recognize synthetic data across multiple modalities, including video, images, 3D, text, and audio.

What is LOKI?

LOKI, a synthetic data detection benchmark jointly proposed by Sun Yat-sen University and the Shanghai AI Lab, aims to comprehensively evaluate the capabilities of large-scale multimodal models (LMMs) in recognizing synthetic data across multiple modalities, including video, images, 3D, text, and audio. It contains over 18,000 questions covering 26 subcategories, employs multi-level annotation, and supports fine-grained anomaly annotation. LOKI tests the model's perception and reasoning abilities, enhancing its interpretability through natural language explanations. By evaluating 22 open-source and 6 closed-source LMMs, LOKI reveals the potential and limitations of these models in synthetic data detection tasks.

LOKI's main functions

  • Multimodal data detection: Evaluate the ability of LMMs to identify synthetic video, image, 3D model, text and audio data.
  • Fine-grained anomaly annotationProvides detailed anomaly annotations to support in-depth analysis and understanding of synthetic data.
  • Multi-level annotationIncludes basic “synthetic or real” tags, suitable for basic problem settings, and more complex anomaly detail selection and interpretation tasks.
  • Comprehensive evaluation frameworkIt supports multiple data format inputs, such as video, images, text, audio, and point clouds, and unifies the APIs of more than 25 mainstream LMMs.
  • Performance ComparisonSupports comparison of different LMMs, including open-source and closed-source models, and expert-synthesized detection models.
  • Explainability testingThe interpretability of LMMs in synthetic data detection tasks is tested by requiring the model to provide natural language explanations.
  • Data diversityIt collected various types of synthetic data, including data from specialized fields such as satellite and medical images, and audio data such as ambient sounds and music.
  • Problem Difficulty LevelThe problem is graded according to human evaluation metrics, and the performance of LMMs is tested at different difficulty levels.
  • Model bias analysisBy calculating the model's bias index, we can analyze the model's bias and tendency in synthetic data detection tasks.

  • Promote AI developmentTo drive the development of more powerful and interpretable synthetic data detection methods to address the challenges posed by AI synthesis technologies.

LOKI's technical principles

  • Data collection and synthesisLOKI collects data from multiple modalities, including videos, images, 3D models, text, and audio, some of which come from public datasets and some generated using state-of-the-art synthetic models.
  • Multimodal evaluation frameworkLOKI proposes a comprehensive multimodal evaluation framework that supports input in various data formats, unifies the APIs of various mainstream LMMs, and evaluates the performance of different models under a unified standard.
  • Model Evaluation and ComparisonThe LOKI benchmark includes evaluations of multiple open-source and closed-source LMMs. By comparing the performance of these models on synthetic data detection tasks, their performance and limitations can be analyzed.
  • Natural Language InterpretationLOKI requires models to provide natural language explanations, enhancing their interpretability. It tests the model's detection capabilities and evaluates the reasons behind its judgments.

LOKI's project address

Application scenarios of LOKI

  • Artificial intelligence security assessmentLOKI can be used to evaluate and improve the security and robustness of AI systems when processing synthetic data, ensuring that AI systems can accurately identify and respond correctly to potential synthetic data attacks.
  • Content moderationLOKI can help detect and filter AI-generated fake news, deepfake videos, or audio on social media, news websites, and other content platforms, protecting users from being misled.
  • Dataset ValidationDuring the training of machine learning models, LOKI can be used to verify the quality and authenticity of the dataset, ensuring that the training data does not contain too much synthetic data and improving the model's generalization ability.
  • Law and complianceIn the legal field, LOKI can help identify and address copyright, privacy, and compliance issues related to synthetic data, such as detecting and preventing the generation and distribution of unauthorized content.
  • Media and EntertainmentIn film, game, and virtual reality production, LOKI can be used to evaluate and improve the quality of synthetic media content, ensuring that the generated content is both realistic and in line with the creator's intent.
Back