debashishpaul / hasib_bhaiya_workshop

What is image classification really?

INTRODUCTION

If you're thinking of building vision systems in the future which can work with the same precision with that of humans, then this tutorial session can help you get started.

This workflow of this session is as follows:

1. An abstract explanation of What/Why/When is Deep learning.
2. Formulating our vision problem assuming we are a computer vision startup!
3. Finding, labelling and cleaning data
4. Picking framework and computer infrastructure
5. Troubleshooting training and ensuring reproducibility
6. Evaluating final model

Welcome to the start of your computer vision journey, probably! In today’s session you’ll set up your deep learning architecture, and train your first image classification model (a convolutional neural network, or CNN), which will learn to distinguish handwritten digit, not the english MNIST database, but bengali handwritten digits for our secret government project on Bengali Hand Writing Recognition System! Let's get started.

Today, we learn and build how to classify bengali handwritten digits. Rather than understanding the mathematical details of how this works(I love Math, or is it meth?), we start by learning the nuts and bolts of how to get a dummy computer to complete the a dead simple task. We will also cover aspect of deep learning such as the painful data preprocessing, data augmentation, transfer learning and also use a technique called ‘fine-tuning’ using pre-trained networks, perhaps the most important skill for any deep learning practitioner

Environment setup

After installing Anaconda, create separate environment for this project Run the following commands:

conda create -n insb python=3.6 anaconda
conda activate insb

Now install the following packages

pip install tensorflow
pip install keras
pip install opencv-python
pip install imgaug

Project directory structure

The project is strucuted in the following format:

# create a folder named dataset and add the numpy arrays that you downloaded from the above link
dataset/
        # training data
        x_train.npy
        y_train.npy
        
        # validation data
        x_val.npy
        y_val.npy

environment_test_notebook.ipynb
workshop_1.ipynb
workshop_2.ipynb