Learning-Deep-Learning

VPGNet: Vanishing Point Guided Network for Lane and Road Marking Detection and Recognition

March 2020

tl;dr: Use VP to guide lane and RSM (road surface marking) detection under adverse weather condition.

The design of catch-all channel dustbin/absence channel to recalibrate softmax heatmap is interesting, and both SuperPoint and VPGNet used the same trick.

Overall impression

This is the first paper on laneline and road marking detection under adverse weather. Under good weather, the performance improvement of VP prediction is not huge. Under adverse weather, it helps quite a lot.

The method requires extensive postprocessing and does not seem to be robust enough for industry use.

VP prediction only helps:

The idea of using VP to guide LLD is similar to LaneNet.

Key ideas

Technical details

Notes