I've worked with image datasets formatted as:
- Class Names as Folders : As the name suggests, images belonging to a specific class are filled into the specific folder with the folder name representing the class.
For example for a dataset to classify Cats vs. Dogs
---all cat images here
---all dogs images here
- Single Folder + Text File : All images are dumped into a single folder - obviously every image file will have a unique name. The key-value pairs of
image_name : class can be stored as rows in a csv file.
--all images heree
- Single Folder with Class in filename : All images can be placed in a single folder with the name of the image having the class label with some changing index value.
Hope this helps!