Hi, I'm Itamar Tsayag

Deep learning pipeline combining YOLOv8, UNet segmentation, and geometric analysis for automated cognitive assessment. Produces standardized 0-10 point CDT scores from hand-drawn clock images for dementia screening.

Deep learning-based automated embryo assessment system integrating segmentation, classification, and temporal reasoning. Multi-modal framework combining U-Net, ResNet, Video Swin Transformers, and Siamese networks for comprehensive embryo evaluation from time-lapse microscopy.

Multi-stage passport photo validation pipeline for ICAO compliance. 98-point landmark detection, geometric registration, and automated cropping at scale.

Automated CT perfusion analysis pipeline for acute stroke. Core and penumbra tissue classification in under 90 seconds to support thrombectomy decisions.

3D body estimation from simple measurements with sub-3mm accuracy. SMPL-based parametric modeling for automated garment template generation.

Novel deep learning architecture leveraging Swin Transformer with window-based self-attention for MRI reconstruction. Features 3-channel temporal processing and multi-domain loss functions.

Master's thesis research on discovering sparse subnetworks at initialization using differentiable gating mechanisms - Achieving high accuracy with extreme sparsity in CNNs, Transformers, and Vision Transformers

Real-time vertical obstacle detection using thermal infrared imagery. Hybrid classical and deep learning approach for low-altitude autonomous navigation.