If the images in the training data set have different sizes and channels and color spaces, will this hinder the classifier? The examples in the documentation seem to suggest that normalization with ConformImages[] on the training and testing data isn't necessary?

  • $\begingroup$ The smaller your training data set, the more important it is to normalize away variations (size, color function, etc.) that are unrelated to the classification task at hand. If you're classifying fruit (where color is valuable), then normalizing the color space is very valuable. If instead you're classifying yoga positions within photographs, then normalizing the color space is unlikely to be of value. Depends upon the classification task. $\endgroup$ – David G. Stork Jan 17 '15 at 2:08
  • $\begingroup$ As @DavidG.Stork points out, the ultimate answer to this is ... it depends. Generally having input which is uniform in features other than those specifically required for discrimination, such as image size, number of channels etc, will be beneficially for the learning algorithm. More detail of your specific application and data might produce more appropriate responses. $\endgroup$ – image_doctor Jan 17 '15 at 13:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.