Hierarchical Multi-Label Object Detection to Analyze Panoramic Dental X-rays

  • Developed a pipeline having a Co-DETR-based tooth detection model and an Efficient-net-based disease classification model with focal loss, intelligent data subset selection and geometric augemntations to counter class imbalance
  • Used Haar-wavelet transform for compression and feature extraction, successfully giving a 50% training time reduction

Download Slides ; Download Report ; Github Repository