Video Analytics using Docker Containers

Video analytics for vehicle using tensorRT and API serving within a docker container. This demonstrates dockerized REST APIs for CNN model using Redis.

Prerequisites

• docker-compose(2.3 version)/Nvidia-docker-compose
• Python 3 #dependency met within docker container
• Keras (with tensorflow-gpu preffered) #dependency met within docker container
• Rest API, Redis/Flask #dependency met within docker container
• Nvidia-tensorRT #dependency met within docker container
• Opencv,Numpy #dependency met within docker container

How to run

We will be using docker-compose-up/down to host the REST API within the docker container.

To start the API, sudo docker-compose up

Use the following curl -X POST command to test the model. You will get a json response relevant to your call. @image/location/image.jpg should be the actual image. Port can be changed in the docker file/server.py. "curl -X POST -F image=@image/location/image.jpg 'http://localhost:5000/predict'"

Models in the Project

You can replace the models with the finetuned models on your dataset. You can swap the model with your model in infer-iva.py.

• Vehicle Make,Color and Type - Resnet model pretrained in tensorRT
• ANPR - WPOD Net for detection of licence plate and yolo(darknet) for character recognition