You are viewing a single comment's thread from:
RE: Computer Vision as a Successor of Human Vision
The last time that I got involved in my thesis for pattern rec, the growing problem for classification seemed to be unsolvable for super huge datasets (too much cross-categorization), and i believe back then i was basing it on k-clustering (or something with that name). How has that progressed so far?
kevinwong, Convolutional neural networks are applied now. Convolutions form features. So trained perceptrone gets class of image. Now there are algorithms, that can recognize 1000 different classes!