korostast/dl-demo-app

Demo Application using Python Visual API

This repo contains:

1. Visual API package is a set of wrapper classes for particular tasks and model architectures, simplifying data preprocess and postprocess as well as routine procedures (model loading, synchronous execution, etc...). An application feeds model class with input data, then the model returns postprocessed output data in user-friendly format. More information see this.

2. demo_application.py - an application that solves one of the following computer vision tasks:

   • classification
   • object_detection
   • segmentation

Usage

To run demo_application next the following steps:

1. Install visual api package:

  python -m venv ml_env
  source ml_env/bin/activate
  python -m pip install -e visual_api_lib/

2. Run demo with needed options:

  python demo_application.py [options]

NOTE: For example was created classification_demo to solve classification task.

Task for students

Main task:

Realize demo application to solve object_detection task. To do it you should perform next steps:

• Choose a DL model for your solution.
• Create model wrapper with pre- and postprocessing operations.
• Choose launcher for model inference. You can choose available launchers from package (onnx or tflite) or create new launcher for another framework.
• Create demo based on updated visual_api (which contains new classes needed to solve object_detection task)

(Optional) Additional tasks

1. Compare performance of your demo on different launchers(backends). Hint: use PerformanceMetrics from common part of visual_api package.
2. Formalize visualization part as separated class or module (maybe as part of visual_api).