How did you go with duplicate detection? (For context, aside from being amazingly generally useful to anyone with non trivially edited images, lots of hotel chains reuse content)
The README explicitly says it's not good for crops, and from the algorithm description it can't be. Or is there another thing called dhash with a different algorithm? https://github.com/benhoyt/dhash. https://github.com/Rayraegah/dhash is similar but horizontal only.