Main object detection

Deep neural network which takes on the input image from any domain and returns main object.

We have trained object detection system which returns locations of different objects, which are presented on the image.

Selection of right object is based on heuristic algorithm which takes in the account such parameters as size of the object, it's type and location.

Problem
Image main object detection with application of state-of-the-art computer vision approaches.
Solution
We used Keras and Tensorflow to develop the object detection model. The body of the neural network is based on the ResNet classifier pretrained on ImageNet.