The project implements a technique called "Transfer Learning" using Imagenet pretrained models VGG19 or ResNet50. Get the pretrained model and verify the accuracy on the CPUs/GPUs. Take a new image data set like for example, fish data-set: http://groups.inf.ed.ac.uk/f4k/GROUNDTRUTH/RECOG/ Use the transfer learning technique to freeze a few layers of the pretrained models and retrain using the new data set Show the accuracy on the new data-set should be roughly equal to or more than original imagenet accuracy. Used TF framework