In question 9 of exam 2021, when you say the region growing algorithm always converges, do you mean that it always finds a solution no matter if it is optimal or not or do you mean it is always optimal? Because as I understand, choosing the right initial seeds is important and affects the segmentation performance
Hi Mouadh,
Yes you understood correctly, "the algorithm converges" doesn't imply that the solution is necessarily optimal. It only means that the algorithm will stop changing at some point, so you can use this criteria as a terminating condition. You could not do that if it was alternating between two solutions at the end, for example.
Corentin
Yes you understood correctly, "the algorithm converges" doesn't imply that the solution is necessarily optimal. It only means that the algorithm will stop changing at some point, so you can use this criteria as a terminating condition. You could not do that if it was alternating between two solutions at the end, for example.
Corentin
Thanks for you answer,
In question 10 of the same exam. Can you explain to me why the region growing does require some examples before processing and why CNNs do not ?
In question 10 of the same exam. Can you explain to me why the region growing does require some examples before processing and why CNNs do not ?
The question states "for each image processed", so that's not necessary for a CNN, it only needs to see labels during training.
For region growing, see the thread with Cyril and Anar (right before your question).
For region growing, see the thread with Cyril and Anar (right before your question).
