Disclaimer: Unfortunately dataset canot be shared publicly due to NDA.

Project was done on small 27+8 microscopy images with a goal of classifying whether each pixel on the image is a human tissue or not. Main challenges were mainly related to sample size, unbalanced classes. two different augmentations pipelines were used with two different models.

• TernausNet (arXiv paper) - U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation implemented from github repository

TernausNet is a modification of the celebrated UNet architecture that is widely used for binary Image Segmentation.

(Network architecure)

Pre-trained encoder speeds up convergence even on the datasets with a different semantic features. Above curve shows validation Jaccard Index (IOU) as a function of epochs for Aerial Imagery

• Vanilla U-Net

The architecture was inspired by arxiv paper and github repository