Tensorflow Count Objects in the image

Question

New to machine learning, looking for some direction on how to get started. The ultimate goal is to be able to train the model to count the number of objects in the image using Tensorflow. My initial focus would be to train a model for counting one particular type of object. So let's say I take coins. I will train the model only for counting coins. Do not worry about creating a common counter for all different types of objects. I just made Google an example of color classification of colors, and I understand the basics of this. So look for clues on how to get started. Is this a classification problem for images and can I use the same logic as flowers ... etc?

Answer 1

Probably the best solution for a coin problem would be to use regression to solve this problem. Annotate 5k images with the number of objects in the scene and run your model on it. Then your model only displays the correct number. (Hope)

Another way is to classify whether the image shows a coin and uses a sliding window approach similar to this: https://arxiv.org/pdf/1312.6229.pdf to classify each window if it shows a coin. Then you count the found areas. It is easier to comment, learn and expand better. But you have the problem of choosing good windows and using the result of these windows in a compressed form.