Detecting a specific number of objects

[styfeng]

Great work! I am wondering if this would work if the user specifies a specific number of objects, e.g. "two chairs". Or will it not work well for ordinal/numerical specifications of objects?

Nvm I guess I can just look at the generated relevance heatmap of "chair" and see if the number of "hot spots" matches the specified number of chairs?

Edit: tried it on the huggingface space and it seems to work decently for numerically-specified numbers of objects :)