I found that the input prompts were different, such as "building" and "house", and the detection results were very different.

[lidapengpeng]

When I run the target detection task, I find that the input prompts are different, such as "building" and "house", and the detection results are very different. Are there any good suggestions to help me with the detection? Do I need to know your open vocabulary glossary?

[rentainhe]

Hello, I think it depends on your needs and the model's actual performance under your scenarios. You may try different prompts in a set of images and compute the performance to determine the final prompt usage.