This project focuses on developing a computer vision solution using Deep Learning for the analysis of embryonic tissues. The solution is implemented using PyTorch and consists of a two-phase model: Segmentation and Profiling.
- Dynamic UNet Architecture from scratch: Designed to process high-resolution microscopy images, enabling precise segmentation of embryonic tissues. Instance segmentation performed on each original image to evaluate the model on complex phenotypic cultures.
-
Custom CNN Model: Built to accommodate variable image resolutions, allowing for the extraction of multiple tissue properties from segmented images.
-
Interactive Annotation Tool: Developed a Annotation tool to facilitate biologists in annotating tissue properties, ensuring high-quality data for model training.