training on new datasets of 100 random images.

I thought of training on 100 random images to check whether own training can be done, but i got this error of **split is empty?**. I have uploaded an image of the error, if the error is because my small dataset can you please help with required changes. Also is it required to modify below:-
1) discard all images that have fewer than 20 or more than 50 annotations.
2)assign 5,000 each to val/test splits and the rest to train.