Why is there such a big difference from your result?

[a3086701440-cmd]

Why did the results on the busi dataset show that the IOU was approximately 10% lower and the F1 score was approximately 20% lower?

[FengheTan9]

Thank you for your interest in our work.

The BUSI dataset often exhibits high performance variance across different data splits. To get a clearer picture of the model's effectiveness, we suggest performing a horizontal comparison of the results w/ LLM4Seg vs. w/o LLM4Seg. This will help isolate the performance gain from our method.

[a3086701440-cmd]

This is the result I obtained without using "LLM4Seg". Why does the actual performance of the pure-blooded 'CMUNeXt' exceed the performance of "LLM4Seg" that I just mentioned?

[FengheTan9]

Deepseek is more stable; you could try it first.