Point t in area loss - Githubissues

wvangansbeke / LaneDetection_End2End

End-to-end Lane Detection for Self-Driving Cars (ICCV 2019 Workshop)

https://arxiv.org/pdf/1902.00293.pdf

Other

643 stars 216 forks source link

Closed wilxy closed 4 years ago

wilxy commented 4 years ago

Hi Wouter @wvangansbeke ,

I have a question when looking at Figure2. The picture is 1280720. Why do you set point t to 0.7, not 896(12800.7)?

Best, Wilxy

wvangansbeke commented 4 years ago

Hi @wilxy,

Point t is set to 0.7 because the grid is already normalized between 0 and 1.

Best, Wouter

wilxy commented 4 years ago

Hi Wouter @wvangansbeke ,

Thank you for your kindly reply. I'm not skilled in pytorch. Could you please point out the normalized code to me?

Best regards, Wilxy

wvangansbeke commented 4 years ago

Hi @wilxy,

The normalization step is extremely easy. Just divide the absolute pixel coordinates (height, width) by the aspect ratio of the image (H, W).

Best, Wouter