Inspiration

As AI is getting popular in any application, I wanted to experiment image classification which can be used in any application. One of the challenges of AI especially image classification is resource intensive in terms of training and validation. Training with 80% dataset and validating the model with the rest 20% dataset.

The idea is to utilize tf libraries for image classification but run it on habana devices, gaudi accelerators.

What it does

The application built is to classify images of flowers. It creates an image classifier using a keras.Sequential model. I train the following flowers dataset from https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz first I train it, compile it, then validate with sunflower with almost 98% accuracy.

How we built it

I used AWS EC2 habana instance to get the environment.

  • Import tf and other libraries
  • Enable habana
  • Download and visualize the data
  • Standardize the data
  • Create the Model
  • Compile the Model
  • Train the Model
  • Visualize the model
  • Address Overfitting
  • Compile the Model
  • Predict

Challenges we ran into

One of my challenges was to identify the correct AMI for EC2, second was loading the habana module.

Accomplishments that we're proud of

Successfully able to classify an image with a high probability score.

What we learned

I have learned how to

  1. Initiate a AWS EC2 habana instance
  2. Loading of habana module in python
  3. Integrate it with tensorflow libraries
  4. Develop, Train, Compile and Predict my model

What's next for AI Image Classification

I wish to extend image classification to object detection and integrate it with my mobile apps.

Built With

Share this project:

Updates