Need Guidance on required folder structure for using VGG (from Lesson1)


Hi Friends, am trying to understand how should I have my folder structure, if I want to use the VGG from Lesson 1, to a problem statement like this - Lets say I need to identify the breed of dog or cat, from the image. Should I need to keep my folder structure(under train, test, valid etc) as dogs -> breed1/breed2 etc and cats -> breed1/breed2 etc? If not, pls guide on how to do the same. And how should the image naming convention be? dog.breed1.1.jpg? Lets say if I want to identify by feature 1 more level deep, should I keep my folder structure as Level -> Level2 -> Level3 and images as Level1.Level2.Level3.1.jpb? Pls advice.

Also if I want to do learning on a data rather than on images, pls let me know how can we go about the same? Can we use the same VGG model? or is there any other model suitable for those kinds?

Pls let me know if you need more details.


(Even Oldridge) #2

Keras treats each folder as it’s own class. As far as I know there is no nesting of classes. So if you want to build a model that predicts breed you need a directory structure that reflects that. Sounds like what you’re looking for is:


You need to keep track after the fact of whether a particular breed is a cat or a dog.


@Even - Thanks. How about the image naming convention? Is there any particular format to keep? Is there any documentation, which I can check on the same?

Also, any suggestion on how to do learning with data and no images?

(Ian) #4

Hi @MLNewbie did you ever get a response for non-data ML?