What is feasible with machine learning and object detection?

Hi, my name is Aeras.

I wanted to find if it’s possible to know which and how many times an object (assuming it was trained) was detected using an open-source tool like TensorFlow/darknet or any other. Then sort those images or video to a folder assigned to them?

I would appreciate it if someone could direct me to into the right direction of tools or readings where I can learn this from.

Thank you

Deeper models like vgg 16, resnet usually perform better in multiclass classification and usually when these models are used and fine tuned they perform better than shallow NNs, however the performance gap diminishes when dealing with binary classification or single/targeted object detection.
To detect a particular object one can go for object detection models however to detect how many times a particular object is present one may go with instance segmentation. Mask rcnn is one such model for instance segmentation.

Tldr; transfer learning can be useful if you the required dataset, instance segmentation for detecting multiple INSTANCES of a particular target (object)

© Copyright 2013-2019 Analytics Vidhya