The current deep learning approaches (i.e. CNN) have used huge training data and they shows better performance (such as in PASCAL VOC challenge). So i am a bit confused that do we really need more training data for better performance or do we better model ??...
Thanks