Introduction

Keras is a high-level Python API to build Neural networks, which has made life easier for people who want to start building Neural networks all the way to researchers.

Use Case

With this project, I want to address a problem that all of us have: **_too many Whatsapp images and no way to sort them. _**As an initial experiment, I made a model that differentiates pictures of people from memes, so that they can be labelled or moved to be stored separately (currently working, hopefully there will be a part 2).

Keras is a BIG library, and thus many of it’s useful functions fly under the radar. In this post, I explore two of such functions:

ImageDataGenerator

ImageDataGenerator is a powerful tool that can be used for image augmentation and feeding these images into our model. The augmentation takes place in memory, and the generators make it very easy to setup training and testing data, without the need of manual labeling of the images

Transfer Learning

Transfer learning is a popular technique, especially while using CNNs for computer vision tasks. In transfer learning, we take a big model that has already been trained for days (even weeks) on a huge dataset, use the low-level features it has learned and fine-tune it to out dataset to obtain a high level of accuracy.

#keras #computer-vision #tensorflow #image-classification #machine-learning