Inquiry regarding the prompt templates used for ADE20K open-vocabulary evaluation

Hi,
In the paper, it is mentioned that SAM3 achieves an impressive 39 mIoU on the ADE20K open-vocabulary segmentation task. I noticed that the paper states SAM3 typically uses short text phrases as prompts, whereas other open-vocabulary models (such as FCCLIP) often use prompt templates like "a photo of a {}."
Could you please clarify if you used similar templates during your testing? If so, would it be possible to share the specific templates you used?
Best regards.