Tensorflow-2-Object-Detection-API-Flask-Application

This is a flask application with tensorflow 2 object detection API deployed. The user hits the endpoint with image data and gets a response which consists of detections with scores, image data with overlay, image size (can be customized).

Prerequisites and Setup

• You need to setup the tensorflow object detection API. Here are some of the great articles that will help you in the process, tutorial1, tutorial2.
• You can the download the required model from tensorflow model zooand specify the saved_model, labelmap path in the main.py file
• client.py is used hit the endpoint. You have to specify the url with endpoint, image_path, output_dir(the dir where you want your image with overlay to be saved. It's optional), in the client.py file.

Run

• first run python main.py to get the app running and then to hit the endpoint with required arguments, run python client.py

Output

The response consists of

• detections with scores
• image data with overlay. This can be decoded back to a numpy array and can be written to disk to visualize the result (provided in client.py)
• image size

Associated Article

https://medium.com/@ms.neerajkrishna/deploy-tensorflow-object-detection-api-on-kubernetes-with-python-flask-and-docker-7a9513dd19e4

References

• https://github.com/tensorflow/models/tree/master/research/object_detection
• https://gist.github.com/kylehounslow/767fb72fde2ebdd010a0bf4242371594