search

zjuluolun / BVMatch

Lidar-based Place Recognition Using Bird's-eye View Images
Apache License 2.0
73 stars 12 forks source link

Confusion about paper and code #3

Open where2go947 opened 7 months ago

where2go947 commented 7 months ago

Thanks for your great work! Howeve, I have some questions about the code implementation, could you explain them to me? Thanks.

  1. According to the formular 5 in the paper: image

In my opinion, the first half corresponds to log_gabor_s: https://github.com/zjuluolun/BVMatch/blob/4010e92aacabd49516af3b3ef9a65170641330a0/src/bvftdescriptors.h#L617-L633

the second half corresponds to spread: https://github.com/zjuluolun/BVMatch/blob/4010e92aacabd49516af3b3ef9a65170641330a0/src/bvftdescriptors.h#L644-L657

But the code of spread seems to be inconsistent with the formula. So how to understand it?

  1. In the paper, you said "the increments are weighted by a Gaussian window function", but actually it is a square calculation. image

https://github.com/zjuluolun/BVMatch/blob/4010e92aacabd49516af3b3ef9a65170641330a0/src/bvftdescriptors.h#L750-L761

  1. Frankly speaking, I can't get the meaning of the following code and can't relate it to the formular in the paper. Could you please explain it in detail? What is the meaning of Mat shang and what is the purpose of calculation of patch based on that?

https://github.com/zjuluolun/BVMatch/blob/4010e92aacabd49516af3b3ef9a65170641330a0/src/bvftdescriptors.h#L801-L807

  1. Lastly, could you explain the main idea of the following code? https://github.com/zjuluolun/BVMatch/blob/4010e92aacabd49516af3b3ef9a65170641330a0/src/bvftdescriptors.h#L828-L854