I am working on my first AI project after Andrew Ng courses. My project is to detect indication on X-Ray pictures.
I would like to know if it is advised or in the legacy to pre-filter pictures before CNN learning ? I found a filter that can help me to see indication in the X-Ray films. I thought it could help the CNN to learn.
Additionally, do you have any model advice for 1 pixel indication identification ? At this time, I am working with Densenet models.
Thanks in advance
Best answer by hamster
What is a filter? Its a NxN kernal right?
What is a CNN? -- A collection of NxN kernals.
So, If your filter is useful, the algorithm might learn it by itself!
But failing that, you could perhaps try using a few specific kernels rather than using random weights in the set-up. This could lead to improved performance, or potentially just getting stuck in sub-optimal points (gradient descent).