should i crop the positive images to contain only the objects to be detected without any background?
U. Dreher
Training involves 'try and error', the training 'effect' done via feedback.
So, if you are at the very beginning - still having to find out what exactly 'makes' a car, images with a neutral background may help.
When it comes to training for the 'real world', the background is exactly what you need to get the training.
2 votes
1 thanks
Walid Ahmed
This question is always in my mind also.
I assume just keep the positive image with object only, during training with positive samples the other pixels are simply ignored.
it is the role of negative images to teach the classifier what is not the object.
what do you think?
0 votes
0 thanks
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning
More Moustafa Elhamshary's questions See All
Similar questions and discussions