Will OpenCV 4.x allow you to create a DNN from scratch, instead of having to import its neurons and weights from a file?
Also, I was wondering why the standard MLP ANN cannot do the cat and dog problem well. I tried to use multiple hidden layers, to no avail. What makes a DNN different than the MLP ANN? I have endeavoured to learn TensorFlow come hell or high water. Hopefully I'm smart enough.
@sjhalayka, If the question is about training deep learning networks in OpenCV, you may find several threads with similar discussions. In example, https://github.com/opencv/opencv/issu.... In short words, we already have a lot of training frameworks. Some of them support distributed training. However still there are challenges how to make deep learning networks run fast in limited resources and OpenCV tries to solve this problem at first. As reducing top hardware requirements (https://github.com/opencv/opencv/pull...) as improving an efficiency using different computational backends and devices (https://github.com/opencv/opencv/wiki...).
Asked: 2018-05-22 10:57:08 -0600
Seen: 439 times
Last updated: May 22 '18
