0
$\begingroup$

I have 3 questions in mind about the neural network

  1. For the best model performance, is it better to train a model only on high resolution images or does it not matter whether the training data includes high-resolution and low-resolution images.
  2. Let us say I would like a model that can detect cats and there are 10 cats I am interested, again, in the best performance. Would it be better if I had 10 classes for each cat or just one class for cats (like just cat) or does it not matter?
  3. Why might a model trained on 2D images not work well when validated by 360 images?
$\endgroup$
0
$\begingroup$
  1. It will depend on the application you are targeting. Certainly, bigger resolution will increase the performance. But if you take that model and apply it in an environment where the resolution is low, then the performance also drops. A good option is if u combine both dataset. That will make your model robust to the resolution change. But that also comes at training cost as you will have higher number of dataset. In general, you should asses your application area and decide a scenario which will enable your model to perform well.

  2. The question is a bit vague to understand.

  3. It comes down to distribution change. Neural networks are very sensitive for distribution changes. To check, plot the distribution of the pixels of both kinds of images and you will notice the difference.

$\endgroup$
1
  • $\begingroup$ Thanks , i edited my second question $\endgroup$
    – learner
    Oct 19 at 21:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.