Tsinghua-MARS-Lab/DenseTNT

The doubt about anchor-free

Xinchengzelin opened this issue · 2 comments

Xinchengzelin commented

I'm confused that the paper said DenseTNT is anchor-free, however, in the code, all the lane boundary point(goals_2D) is the model input, Is this contradictory？

GentleSmile commented

The meaning of anchor here is the same as in object detection, where anchors are used to regress bounding boxes. TNT uses anchors to regress final goals, which is anchor-based. In DenseTNT, no regression is needed.

Xinchengzelin commented

e used to regress bounding boxes. TNT uses anchors to regress final goals, which is anchor-based. In DenseTNT, no regression is needed.

OK, Got it