CODEWITHSUNDEEP
opencvcomputer-visiondlib
user3282276
user3282276

Reputation: 3804

Does the type of image matter when training an object detector?

I was wondering if the type of photo used to train an object detector makes a difference, I can't seem to find anything about this online. I am using opencv and dlib if that makes a difference but I am interested in a more general answer if possible.

Am I correct in assuming that lossless file formats would be better than lossey formats? And if training for an object jpg would be better than png as pngs are optimized for text and graphs?

Upvotes: 1

Views: 68

Answers (1)

Davis King
Davis King

Reputation: 4791

As long as the compression doesn't introduce noticeable artifacts it generally won't matter. Also, many real world computer vision systems need to deal with video or images acquired from less than ideal sources. So you usually shouldn't assume you will get super high quality images anyway.

Upvotes: 2

Related Questions