Detection of AI-Generated Images

Inter IIT Tech Meet 13.0 - Classify AI-generated images, identify artifacts.

  • Implemented a BNN with Fourier Magnitude and Local Binary Pattern channels for classification
  • Benchmarked Static & Dynamic Quantization (PTQ & QAT) in PyTorch for real-time performance
  • Used Jina-Clip-V2 and ResNets for artifact identification; fine-tuned Ovis1.6 for explanation generation
  • Results: 2.6× speedup (80ms → 30ms) with <1% accuracy drop using Dynamic FX Mode PTQ; 97.2% accuracy, 30ms CPU inference