Serverless Image Classifier

Build an end-to-end deep learning model to classify real-world images using TensorFlow, Docker, AWS Lambda and API Gateway

Photo by Benjamin Rascoe on Unsplash

Documentation

• GitHub Repository

• Towards Data Science

Summary

• Developed deep learning models to classify 10-class images in Python and TensorFlow for a clothing e-commerce business.

• Optimized the convolutional neural networks with transfer learning and image augmentation to achieve a good performance and prevent overfitting, leading to a 91% accuracy with 3.04% gap between the training and test accuracies.

• Optimized the model for deployment in mobile and edge devices using TensorFlow Lite, resulting in a 75% reduction in size for efficiency.

• Deployed the model locally using Docker.

• Built a public facing API using AWS Lambda and API Gateway.

• The following is the URL to make the HTTP API call:

https://xw2bv0y8mb.execute-api.us-east-1.amazonaws.com/test/predict

• Image sample data to be sent as a POST request:

{
    "url":"https://tinyurl.com/clothes-t-shirt" 
}

Visualization