I am trying to build an application to count objects in satellite imagery e.g ships or cars outside a supermarket.
Now, the labeled dataset that I have contains images of 80x80 px windows of imagery and labels if the object is present in the image or not e.g ship or no-ship. I was able to build a good classifier using the dataset.
Now, to count the no. of objects, I can divide the image into windows and count the no. of windows with the object. I want to know if there are better alternatives.
What do you guys think? Let me know should you need more details on the problem.
I’ve successfully trained OD models on sat/aerial imagery and currently working on improving accuracy. PM me if you’re interested in partnering up to tackle this problem.