August 2018
Intermediate to advanced
272 pages
7h 2m
English
R-CNN is still computationally expensive, as you have to run a CNN on around 2,000 individual region proposals. As a result, it is very slow for both training and testing. The CNN classifier relies on the fixed number of rectangular candidate windows produced by selective search for detection. This method is not the fastest around and as proposal regions are not learned from the training data, they are probably not optimal for the task.
Read now
Unlock full access