csuhan / s2anet

Official code of the paper "Align Deep Features for Oriented Object Detection"
376 stars 90 forks source link

谢谢!!急求!请问计算tensor(N,5)的两个旋转框的IOU的源码在哪个文件下 #62

Closed yzk-lab closed 3 years ago

yzk-lab commented 3 years ago

你好,我现在在处理IOU,但是在mmdet,core或者bbox下都只看到了处理(N,4)张量的IOU的方法,我想知道在哪里有计算旋转框IOU的代码?这个s2anet项目里是否有提供?谢谢!

csuhan commented 3 years ago

Please refer to Here

yzk-lab commented 3 years ago

thanks for your reply.I have know how to get IOU.there exists another problem.i want to replace the smoothl1loss with iou loss .but i print the bbox_pred and target ,i found the target is a tensor with 0.so the IOU bwtween the target and pred is 0,therefore my iou loss is NAN.if i want to use the iou ,how can i output the iou between the anchor and gt_bbox?i cannot get the IOU between the pred and target.hope for your reply.this problem have puzzle me for a long time.

yzk-lab commented 3 years ago

i used your original s2anet to replace the smooth-L1loss with iou loss .it failed

csuhan commented 3 years ago

To use IoU as the loss metrics, I suggest the PIoU loss: https://github.com/clobotics/piou

yzk-lab commented 3 years ago

are there any solution with IOU by your s2anet?

------------------ 原始邮件 ------------------ 发件人: "csuhan/s2anet" @.>; 发送时间: 2021年3月14日(星期天) 晚上7:08 @.>; @.**@.>; 主题: Re: [csuhan/s2anet] 谢谢!!急求!请问计算tensor(N,5)的两个旋转框的IOU的源码在哪个文件下 (#62)

To use IoU as the loss metrics, I suggest the PIoU loss: https://github.com/clobotics/piou

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

yzk-lab commented 3 years ago

sorry to bother you.can you tell me input parameters (the pred and target )in loss function where i can touch,it seems i can not understand the target why the anchor_target is a tensor only include zero.i need to know the meaning of pred and target.

------------------ 原始邮件 ------------------ 发件人: "csuhan/s2anet" @.>; 发送时间: 2021年3月14日(星期天) 晚上7:08 @.>; @.**@.>; 主题: Re: [csuhan/s2anet] 谢谢!!急求!请问计算tensor(N,5)的两个旋转框的IOU的源码在哪个文件下 (#62)

To use IoU as the loss metrics, I suggest the PIoU loss: https://github.com/clobotics/piou

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.