I've only worked with ML with .csv formats. I've worked with image formats too but only premade imagesets (MNIST,etc). If I were to create an imageset from scratch, how are the class labels typically formated? Would I have to manually title the image of a jpeg?

Best, Jeremy

  • $\begingroup$ also, jpeg images cannot be the same title, so wouldn't the ML program recognize every image as its own class if all titles different? $\endgroup$ Mar 28, 2015 at 22:36

1 Answer 1


Looking at existing challenges around and their data format (for example http://www.kaggle.com/c/datasciencebowl/data) I would say put the images in a folder per class. You can use the file names for the index.

  • $\begingroup$ This is right. I asked someone at Ersatz Labs too. Same answer: group class by folder. Thanks. $\endgroup$ Mar 29, 2015 at 20:16
  • $\begingroup$ by the way, for that Kaggle link. That is a TON of classes for only 30K images. Interesting. Thanks for sharing that. $\endgroup$ Mar 29, 2015 at 20:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.