engcang / tkdnn-ros

(ROS) YOLOv3, v4, v7 detection + Shelfnet semantic segmentation with TensorRT, utilizing tkDNN
BSD 3-Clause "New" or "Revised" License
12 stars 3 forks source link

the screen is full of bounding boxes #1

Open ningjingnn opened 2 years ago

ningjingnn commented 2 years ago

IMG_0478 IMG_0481 When I use the imshow function, the screen is full of bounding boxes, do you know what the reason is?I've only made some changes to your code.

engcang commented 2 years ago

@ningjingnn Hi.

When initializing the tkDNN handler, did you pass the propoer confidence threshold and class number? https://github.com/engcang/tkdnn-ros/blob/master/include/main.h#L94

ningjingnn commented 2 years ago

Hi,Sorry to bother you again. My confidence threshold is already high, set to 0.7.And my class number was right.My same weight is fine in TKDNN, but testing under ROS is problematic.

------------------ 原始邮件 ------------------ 发件人: "engcang/tkdnn-ros" @.>; 发送时间: 2022年7月19日(星期二) 下午4:13 @.>; @.**@.>; 主题: Re: [engcang/tkdnn-ros] the screen is full of bounding boxes (Issue #1)

@ningjingnn Hi.

When initializing the tkDNN handler, did you pass the propoer confidence threshold and class number? https://github.com/engcang/tkdnn-ros/blob/master/include/main.h#L94

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

engcang commented 2 years ago

@ningjingnn Hi. Can you show/give me full code of you. I would reproduce the error.

ningjingnn commented 2 years ago

  Hi,My code is in the attachment, and I made some changes on your main.h. Because when I use main.h's code, I can't subscribe to the camera stream. According to the idea of main.h, I wrote a code similar to main.h in the attachment, which can be subscribed to the camera stream normally, but there is a problem with the detected video.   Thank you very much for your time and energy!

------------------ 原始邮件 ------------------ 发件人: "engcang/tkdnn-ros" @.>; 发送时间: 2022年7月19日(星期二) 晚上11:01 @.>; @.**@.>; 主题: Re: [engcang/tkdnn-ros] the screen is full of bounding boxes (Issue #1)

@ningjingnn Hi. Can you show/give me full code of you. I would reproduce the error.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

ningjingnn commented 2 years ago

The attachment is my weight file.

------------------ 原始邮件 ------------------ 发件人: "engcang/tkdnn-ros" @.>; 发送时间: 2022年7月19日(星期二) 晚上11:01 @.>; @.**@.>; 主题: Re: [engcang/tkdnn-ros] the screen is full of bounding boxes (Issue #1)

@ningjingnn Hi. Can you show/give me full code of you. I would reproduce the error.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

从QQ邮箱发来的超大附件

yolo4tiny_fp32.rt (61.12M, 2022年08月19日 10:36 到期)进入下载页面:http://mail.qq.com/cgi-bin/ftnExs_download?t=exs_ftn_download&k=233634384471199d65c927334762054e5105010c515a04044806010e564f03045103195a5104544c5204570e53525305030f510b617337180a5a5b0c150b59183a50440b534c4515650b&code=e648ab7a

engcang commented 2 years ago

@ningjingnn Hi. Sorry for late reply. I can find your .rt file but I cannot find your code attached. If you upload the full code of yours, I will debug it!

Apart from that, I tested my code with TensorRT version 8 today and it works well.