search

JKBox / YOLOv3-quadrangle

YOLOv3 detector with quadrangle in PyTorch
88 stars 29 forks source link

The way of coordinates regression repo used in whether is same with Deep Matching Prior Network used in #5

Closed ygean closed 4 years ago

ygean commented 5 years ago

@JKBox link

JKBox commented 5 years ago

@JKBox link

yes, it's roughly the same

ygean commented 5 years ago

@JKBox Hi, thank you for your reply. Is there any blog you write to explain how it works properly for this repository? I have been read your codes for several days, actually, in the original paper, computing IoU of Polygons use Shared Monte-Carlo method, but from your codes I got that you used sharply to compute the IoU, So I have a question that, sharply module does that job may slower than Shared Monte-Carlo, am I right? sharply can not process cuda tensor actually. So I want to contribute this repository to implement `Shared Monte-Carlo, could you help me or give some suggestions? Sincerely, best wishes~

JKBox commented 5 years ago

@JKBox Hi, thank you for your reply. Is there any blog you write to explain how it works properly for this repository? I have been read your codes for several days, actually, in the original paper, computing IoU of Polygons use Shared Monte-Carlo method, but from your codes I got that you used sharply to compute the IoU, So I have a question that, sharply module does that job may slower than Shared Monte-Carlo, am I right? sharply can not process cuda tensor actually. So I want to contribute this repository to implement `Shared Monte-Carlo, could you help me or give some suggestions? Sincerely, best wishes~

Thank you for your interest. I implemented Shapely module just for convenience, and it's indeed much slower than Shared Monte-Carlo method. Sorry that I can't give you any suggestions cause I don't know it well. You can try it, good luck.

Jinglei5 commented 4 years ago

@zhouyuangan hi~Did you find the implementation of Shared Monte-Carlo yet? Could you share the repository or link? Thanks a lot!

ygean commented 4 years ago

@zhouyuangan hi~Did you find the implementation of Shared Monte-Carlo yet? Could you share the repository or link? Thanks a lot!

Sorry, I had sent an email to author but no response I got, and I gave up my plan. So I'd like to compute IoU and overlaps between rotated boxes. It's easier, thanks.