Currently, most examples when querying for a specific object/class type make use of match() or filter_labels(), to the point where even the you can use these and only these label classes: in prompts/dataset_view_generator_prefix.txt is ignored and classes that aren't listed but are in the prompt will be used.
This leads to empty views since filtering on a label class that doesn't exist won't result in anything interesting, even if a text similarity brain run exists which could give very interesting results.
For example, searching the coco_gpt_demo dataset with images with a camel in them should result in a sort by similarity stage, but tried a match instead.
Currently, most examples when querying for a specific object/class type make use of
match()orfilter_labels(), to the point where even theyou can use these and only these label classes:inprompts/dataset_view_generator_prefix.txtis ignored and classes that aren't listed but are in the prompt will be used.This leads to empty views since filtering on a label class that doesn't exist won't result in anything interesting, even if a text similarity brain run exists which could give very interesting results.
For example, searching the
coco_gpt_demodataset withimages with a camel in themshould result in a sort by similarity stage, but tried a match instead.