Hi,
I’m currently building a dataset and I’m facing something weird :
When I split my data (train/valid/test) with the default 70%/20%/10% split, after the augmentation step, the split is not the same anymore.
It seems like it’s only increasing the data in the training set.
I don’t know if it’s how it’s supposed to work, hope someone can help/explain.
Hi there! This is because the split works on your original set of source images. After you create that split, we augment only images in the train set. If valid or test images were augmented, this would hurt model performance.
