This project (in collaboration with Tomer Gofman) is a Computer vision algorithm pipeline based on task transfer of pre-trained ResNetV50 on ImageNet dataset, from multiclass classification problem (1000 classes) to binary classification task trained and tuned on about 150 images. The pipeline includes pre-processing including data augmentation, various hyper-parameter tuning, as-well-as architecture changes and improvements, and a results report. The code was developed on Jupyer environment in python.
classification accuracy for the binary problem - about 90%.