IVideoSafetyModule<T>

Interfaces ยท AiDotNet.Interfaces

Interface for safety modules that operate on video content.

Implements: ISafetyModule<T>

Known implementations: VideoSafetyModuleBase<T>

For Beginners

Video safety modules check video content for harmful material. They can analyze individual frames (like image safety) and also detect temporal inconsistencies that reveal deepfake manipulation โ€” things like unnatural blinking, facial warping between frames, or audio-visual mismatches.

How It Works

Video safety modules analyze video frames and their temporal relationships for safety risks such as harmful visual content, deepfake videos, and policy-violating material.

References:

  • Spatio-temporal consistency exploitation for deepfake video detection (2025)
  • NACO: Self-supervised natural consistency for face forgery detection (ECCV 2024)
  • Generalizable deepfake detection across benchmarks (CVPR 2025)

Methods

Method Summary
IReadOnlyList<SafetyFinding> EvaluateVideo(IReadOnlyList<Tensor<T>> frames, double frameRate) Evaluates the given video frames for safety and returns any findings.