for now we are concatenating training annotation in one file, assuming that the training script will be able to split training data into train/validation sets.
Try to use the current implementation, but in case that yolo v5 does not support this feature, we should implement this as an improvement to yolo v3 to v5 conversion script