AISaturdayLagos: Classification of Nigeria Currency Notes Using Fastai Framework
We held AI Saturdays Lagos Week 5 on 3rd Feb, 2018. We went through Fast.ai lesson 4, the lecture worked through the the previous lesson on tabular/structured, time-series data, and we learnt about how to avoid overfitting by using dropout regularization. We were then introduce to natural language processing with recurrent neural networks, and started work on a language model. After this session, we took an hour break which was used for lunch plus class interaction. After lunch, we saw Lecture 3 of CNN, which was about Introduction to Computational Neural Networks, how the covnets are built on top of one another. We later held a deep learning theory session where we talked about the curse of dimensionality, why deep networks are better than shallow networks. Project group discussion proceeded afterwards.
Kenechi Franklin Dukor did some experiment with the Fastai framework to classify different Nigeria currency notes. Here’s an excerpt of his medium article.
The fast.ai course has proven to be one of the fastest learning option for beginners who want to build deep neural networks with good results in image classification. The course begins with coding and gradually goes dipper into the theory of deep-learning (i.e. bottom-top learning method).
After day-three at the Saturday-AI class by Nurture.AI in Lagos Nigeria, I decided to try out the fast.ai library with the procedures laid-out in the course; a simple image classification task using the Nigerian Naira as subject.
I started off by gathering random images of the naira denominations (N5, N10, N20 … N500, N1000).
After gathering sufficient images from google search engine, I then categorized the images in to ‘train’, ‘test’, and ‘valid’ folders.
Link to my dataset: My Data-Set.
For this practice, I had 32 training set and 15 valid set for each category of the Naira currency. I also had 57 test set for the practice.
The resnet34 algorithm was used for the classification with a learning rate of 0.01 and 3 epochs.
…
…
…
Thank you Kenechi Franklin Dukor for a beautifully written article.
AISaturdayLagos wouldn’t have happened without my fellow ambassador Azeez Oluwafemi, our Partners FB Dev Circle Lagos, Vesper.ng and Intel.
A big Thanks to Nurture.AI for this amazing opportunity.
Also read how AI Saturdays is Bringing the World Together with AI
See you next week 😎.
