Some instance may have multi-label during prediction?

hustvl / MapTR

[ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction

MIT License

982 stars 150 forks source link

Some instance may have multi-label during prediction? #155

Open drilistbox opened 5 months ago

drilistbox commented 5 months ago

791274cead02757b98d35a3c89642f3

the 22th instance is predicted as predestrian crossing, lane divider and road boundaries with different scores. Through a score threshold is used to select the effective target.

I just confuse why not directly choose label with top-1 scores for each instance. Is there any motivation？