CSCI8955-Advanced-Data-Analytics

Scalable Deep Learning Object Detection & Model Comparison

• This project has following features :
  1. AWS serverless architecture
  2. Lambda function integrated with Keras, Tensorflow and Python Libraries.
  3. User Interface for uploading an image and evaluate the result on different models.
• We created a pipeline using AWS components for an end-to-end application which has the advantage of the scalability and extendable to any deep learning model. Presently it is supported for Faster-RCNN and SSD.
• Important files to look for implementation:
  • /lambda/frcnn
    • frcnn-detection/lambda_function.py
    • frcnn-lambda/service.py
    • frcnn-preprocessing/lambda_function.py
  • /lambda/ssd
    • ssd-detection/lambda_function.py
    • ssd-lambda/lambda_function.py
    • ssd-preprocessing/lambda_function.py

Requirements:

- Python 2.7
- Keras 1.2.1 and Keras 2.0.2
- Tensorflow 1.2.1
- OpenCv 3.1
- Numpy
- Scikit-image
- AWS DynamoDB
- AWS API Gateway
- AWS Lambda functions
- NodeJs 6.1.1

References:

• https://github.com/yhenon/keras-frcnn
• https://github.com/rykov8/ssd_keras