Skip to content

Ideas for 2.0

  • Add Sanity check on images (it's an image, not a screenshot of a social media post, with an image inside)
  • T-sne for low-dim visualization of CLIP embeddings to assess coverage of synthetic/real datasets (dataset's biases can be partially uncovered using CLIP)
  • Extend tasks : deepfake classification & pixel-based forensics
  • Perform high-level analysis on results

Add more generation models

Edited by Raphaël Couronné