Is it possible to add more than 11 convolutional networks, for example 13*13(filters)? what is happening ? Is it possible to obtain more general features that may be useful for images that contain more details?
The answers to your questions are “yes”, “this is covered in the lectures”, and “yes”.
Thank you for the quick answer, but most of the well-known techniques, such as Resent, use a maximum number of 11*11? My question was about whether anyone has used or tried to use anything above that
What part of the course are you specifically referring to?
If it’s from a lecture, please give the title and a time mark.
My question was general about the possibility of using convolutional networks using high-content filters for high-resolution images. Thank you again
I cannot say that no one has never used more than any specific number of filters.
This is an experimental science: you can try using larger filters like 13 x 13 or 15 x 15 and see whether that gives you better results. But if you survey the existing published models that researchers have developed to solve various kinds of image processing problems and find that the largest filters you see are 11 x 11, that probably tells you that larger ones have been tried and didn’t contribute anything. But maybe those decisions were made at a time when computational resources were more limited and if you tried it now with current generation GPUs and modern huge datasets, you’d find that it makes a difference. Or maybe you’re trying to detect something in images that has a larger scale and in your particular case the larger filters would help. The way to find out would be to try it. If you do run experiments like this, it would be interesting to hear what you learn.