Elements of Deep Learning for Computer Vision by Bharat Sikka

Author:Bharat Sikka [Sikka, Bharat]
Language: eng
Format: epub
Publisher: BPB Publications
Published: 2021-07-15T00:00:00+00:00

A simple formula to add padding in such a way is given below:

P: Padding size

K: Kernel size or frame size

Transfer learning

The final topic to discuss before we move on to the implementations of a CNN classifier program is transfer learning. We have already seen and understood earlier that in deep learning we are working with a huge number of layers, parameters, and our computations increase as we work on image data that includes multi-dimensional arrays. For training neural networks, it might take days or weeks on multiple GPUs on huge datasets, hence we need to have a way to reduce this time complexity.

Specifically for such issues, we use the strategy of transfer learning, where in, we want to train a network with less amount of data and reduced time. We use the weights of a pre-trained network and add it to the second last layer of our network and re-train it with our data. This gives us an acceptable performance network with much less data and time required to train the full network again.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.