How to use EfficientDet for semantic segmentation?

In the EfficientDet paper, section 5.2. 5.2. EfficientDet for Semantic Segmentation, the authors say

we modify our EfficientDet model to keep feature level $$\{P2, P3, ..., P7\}$$ in BiFPN, but only use $$P2$$ for the final per-pixel classification

How to use only $$P2$$ for semantic segmentation purposes in this case?