A convolutional neural network built on pretrained models such as ResNet50 by transfer learning to classify images of feet into three categories - Flat, Medium Arc, High Arc. Dataset was provided by a company called MyFootDoctor in a hackathon to automate the process of analyzing feet defects. Accuracy of approx 80% was achieved w/biased dataset.
A web interface with react js was also built to test the model in real time. Along with contour detection by grayscaling the images using computer vision.
Libraries used: Keras & Tensorflow