0
$\begingroup$

I am working through ideas for a computer vision problem, and most of the examples I see online use extremely small images, often 32x32 pixels.

Does it make sense to try to train a neural network on images of 1024x768 pixels, or it that out of the realm of current computing power. Using a GPU cluster in AWS is in play, I'd just like to know if it's worth the time to try out that path.

What are some other techniques for training models on HD image data?

$\endgroup$
1
$\begingroup$

There is no harm training a neural network model on large images if you have the computational power. However, even huge computational resources will take a very long time to train a model on high-dimensional inputs. My suggestion is to reduce the size of your images, if possible, and use convolutional neural networks as your model instead of fully-connected neural networks. Convolutional neural networks have much less number of parameters compared to fully-connected models while inherently reducing the size of images with pooling layers and yield much better results on visiom problems.

$\endgroup$

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.