search

yuliangguo / Pytorch_Generalized_3D_Lane_Detection

[ECCV 2020] Official PyTorch implementation of 'Gen-LaneNet: a generalized and scalable approach for 3D lane detection'
307 stars 78 forks source link

A question about estimate z for 3D lane #4

Closed Monkey-Hu closed 4 years ago

Monkey-Hu commented 4 years ago

Thanks for your amazing work. I have read Part 3.1 and I just cannot understand the process of estimating z for 3D lane. After the first and second trasformation，I think there should only be red curves on the following image. So where do these blue curves come from? Thank you very much for answering my questions . 1596187663(1)

maxenceliu commented 4 years ago

both red and blue lines are ground truth, which use different projection method. reds are IPM projection. blues are (x,y,z）-->(x,y,0)

Monkey-Hu commented 4 years ago

both red and blue lines are ground truth, which use different projection method. reds are IPM projection. blues are (x,y,z）-->(x,y,0)

that makes sense. thanks for help~

yuliangguo commented 4 years ago

@Monkey-Hu Thank you for the question. @maxenceliu Thank you for the correct answer.

voldemortX commented 3 years ago

@yuliangguo I've read your paper and @maxenceliu's explanations, but I'm not very good at geometry, so I still can't fully understand this. If reds are got from IPM, shouldn't lanes be parallel already? Why do we need to move them to be parallel? My main question is: how is it possible to ascertain heights with 2D segmentation map on the captured image as input? Could you guys elaborate a little so I could understand this better.