Correct way to implement a binary classifier?

Question

Neural networks

I want to practice in convolutional neural networks and to start making a binary image classifier. Those. roughly speaking - hotdog / not_hotdog. As I understand at the moment, there are several ways to implement this:
1. Create a grid with 2 neurons at the output, the first neuron is activated if there is a hot dog, the second - respectively, in other cases. We make two groups of images for each case. We train.
2. We create a grid with 1 output neuron, which is activated only if there is a hot dog. Otherwise, the output is 0. The training sample contains only hot dogs.
Tell me, do both options have the right to exist in practice, which one is better, maybe there are more options?

Answer 1

In any case, the training sample should contain both positive and negative examples.
Doing one neuron with a sigmoid at the end or two with a softmax - in general, it's all the same.