Name of validation folder for classification training (Export: Folder Structure)

Hi,

when exporting an image classification dataset from Roboflow as Folder Structure format with train/validation splits, the folder which contains validation images is automatically named valid. However, YOLOv5-cls training expects the validation folder to be named val (see here) and throws an error if there is only a valid folder (and no test split). I checked all Roboflow notebooks for classification training and in none of the training procedures a valid folder is actually used for validation.

If the default name for the validation split when exporting a dataset as Folder Structure format would be changed from valid to val, YOLOv5-cls classification training would work as expected, when no test split was created.

Thanks!

Start by tracing the Folder Structure exporter and compare its validation-directory naming with YOLOv5's classify/train.py expectation. Done means an export with train and validation splits but no test split produces a val directory that YOLOv5 classification training can use.