How to do character recognition using neural networks with your own set of pictures?

Question

Neural networks

Hello,
I have already read a bunch of articles on machine learning, on different frameworks for it, but I still don’t understand anything :)
Or rather, an example of character recognition from the MNIST database is given everywhere, but it’s not clear how to make this example by hand.
Can someone suggest a simple example. Here we are uploading a bunch of our pictures for training, testing, checking.
The language and the framework are unimportant, the main thing for me is to capture the essence.
And at the same time here is a little stupid question.
Let's say I have a set of pictures with cars. I need to know if it's broken or not. Do I need to somehow process the image for this task so that the neural network can distinguish between these two images? Or will she herself deduce a pattern due to some common repeating patterns?

Answer 1

The neural network does not know how to "see" how a person does it. She needs to be "suggested" what is good and what is bad.
On reflection: when the car is intact, then its body shape is even, the lines are solid, in a single fragment.
So it is necessary to formalize these features in the form of numbers, encode the input data, and the neural network will already classify according to these features.

Answer 2

MNIST recognition can be done on different architectures. But convolutional neural networks are now used to recognize complex images. With proper training, they can detect various patterns in the image (such as a crumpled body, broken glass) and draw a conclusion about the state of the car.