CV use cases for multi-scale CNNs [Research]

I am looking for computer vision tasks where the use of multi-scale feature extractors is crucial for performance. Links to studies establishing the importance of multi-scale features for any CV task would be much appreciated!