Hello, please help. Using MLAgents, I try to tune the neural network to determine the pictures. I create pictures of 30 pieces of the same suit and save them to a .bin file (an array of bool of size 1024 is saved, where the filled pixel is true) and feed this base to the neural network, types of picture 22. The neural network is trained, but it does not accurately determine the pictures, makes mistakes and confuses pictures … Here is the code https://pastebin.com/ubx5mqAr. Am I doing everything right? And yet, if the neural network has not determined the pattern, then what to expect on the output neurons? or in general, how do you know that the neural network has not detected the drawing? Sorry for my English.
It looks like you’re trying to tackle an image recognition problem.
This kind of problem can be solved pretty well by supervised learning in general.
ML-Agents is a toolkit that uses Reinforcement Learning to train behaviors for game agent so it might not be the best tool for your case. You might want to look for general image recognitions tools.
Another point is you’re feeding an image input as a vector and it will lose the information of relative position between each pixel and will affect the training performance.
Understood thanks. So I need to use the CameraSensor and supervised learning?
1 Like
Following this thread!!.. as I only find things related to AR recognition, and not screen images recognition