Simplified - Building an object detection app

Object Detection is an important sub area of Computer Vision tasks, which has application in almost every domain - Medicine, Drone Analytics, Agriculture, Road safety and many more.

This is one of the difficult tasks in Deep Learning, because there are dual objectives involved - What and Where (Classification and Localization). Traditionally, building a system having object detection capability has been quite challenging because of Modelling and Productionizing a model for end users.

Thanks to the recent interest and a vibrant community of AI researchers and Product designers, we have enough tools available to build a system in a weeks time, provided the data is available. And, I am not talking of a rudimentary system, I am talking of a system, which has state of the art capabilities.

I have built a Pneumonia Detection App which is an object detection which is available on Github. This project uses SIIM dataset which has got images available in dicom format and information available around four conditions. Information is given in a bounding box format and associated label is provided with the bounding box.

The project addresses the modelling and consumption of the model using YOLOv5 (You only look once) and Gradio framework and running this project as per the instruction provided yields a nice user interface supported by an object detection model. This interface can be used to examine a chest X ray and look for conditions associated with them along with the co-ordinates. Here is an example.

This interface can also be run in debug or observation mode. An example is here.

This is an open source project and you can also go ahead and build your own system with this. Just clone the project and follow the instructions and you will be able to build your own object detection system in no time. Happy Detecting!