Open wujiang0156 opened 4 months ago
"I haven't extensively tested the set of 9 here, but in reality, you can try other parameters for comparative experiments. However, I don't think it's necessary because this set of 9 is not the focus of the article, and discovering the optimal N is not an innovative research approach. You should try to design a better scheme for text description of variations."
You retrieved the top 9 high-confidence categories to assemble text prompts for foreground and background features, resulting in better training outcomes. Why choose 9? What would happen if you chose a different number of confidence levels?
Looking forward to your response.