search

HaozheQi / P2B

P2B: Point-to-Box Network for 3D Object Tracking in Point Clouds
189 stars 36 forks source link

seed targetness label #1

Closed gongshichina closed 4 years ago

gongshichina commented 4 years ago

https://github.com/HaozheQi/P2B/blob/9015b9161db594d4650bfacb8aea49c36b7befc9/kitty_utils.py#L125

Hello, thanks for your good job. I have a question about this line in your code:

Why we can just take the first 128 numbers in the target mask as seed targetness label?

HaozheQi commented 4 years ago

Hi, @gongshichina ,

Thanks for your attention to our work.Given an input point cloud from search area, we normalize the component points into 1024 ones with random abandoning or duplicating. The result points are fed into three-layer PointNet++ with 8 times down-sampling and yielding 128 seeds. The downsampling is actually a random process. Since the input points have been randomized, here we sample the first 128 points as final seeds. This explains the number selection in your question.