My 3 months with Computer Vision — Part 1

Computer Vision

Recently, I started looking into Computer Vision and how to approach these kind of projects. In the past 4 months, I have gone through Coursera, Udacity and Udemy courses. Here is the summary of how to approach any Computer Vision Project -

1. Do EDA to understand what is to be predicted.
2. Decide on the model to run: we will run 2 projects with our custom model and 2 with Transfer Learning
3. Do data augmentation to generalize your model to work on unseen data.
4. Finally, evaluate your model, freeze it and deploy.

All projects are done in Keras. We will be doing 4 projects:

1. MNIST : In the MNIST we will work with (28, 28) images and custom model to get an idea of the different layers there are in Keras.
2. Dog Vs Cat: In this we will be differentiating between images of cats and dogs. We will use custom model again and show you don’t need Transfer Learning for smaller learnings.
3. Stanford Dog Breed: This is dataset from Stanford. It contains images of dogs of 120 different breeds. We will use transfer learning for this.
4. Udacity Facial Keypoints: This is Project 1 from Udacity in their computer vision Course. We will again use Transfer Learning. Along with this, we will also be writing our own data augmentation generator.

Let’s start by understanding Neural Networks.