How can I label images with numbers for the purposes of machine learning?

I am new to MATLAB.

For example, if i have a series of images of jars with m&ms,

and i have the true number of visible m&ms per jar.

How do i label these so as to train a network

capable of taking an image as input

and yielding number of m&ms as output?

Answer 1

[Peng Li]

One solution is to save images with the same number of mms in one folder and name that folder with the number of mms. You can load these folders as an imagedatastore. It has property to specify classes as folder names.

I have a set of deep learning scrips with matlab that you may be interested in.

Check

https://github.com/pliphd/neural-network-and-deep-learning-with-MATLAB

[MATMania]

thanks Peng! but if i do that, will it label them with something that it considers a number - ie, "14" - or will it name it a string of text ie "1" followed by "4"?

another way to ask the same thing: suppose the training set has images with 100, 110, 140, and 180 mms. then if if a test case has 165 - would the approach you suggest lead to "classification" as "165", or as "180", because it doesnt have a "165" category?

i am new to this so apologies if the question is unclear.

[Peng Li]

This is tricky for me. As anyway your model is predicting several classes, it only throws you categories you have defined.

Instead of doing a classification, will regression be working for you case? All other layers the same just the final classification layer being replaced with a regression layer.

[MATMania]

thanks Peng. regression would be great i think. could you refer me to documentation thats good for regressing images to a integer value?

[Peng Li]

https://www.mathworks.com/help/deeplearning/ug/train-a-convolutional-neural-network-for-regression.html

see this example in matlab documentation.