mAP of yolov3-tiny on VOC

[developer0hye]

Hi, yjh! It's been a while.

Have you ever measured the mAP of yolov3-tiny on VOC?

[yjh0410]

Hi! Long time no see! The mAP on VOC,, emmm..... I remember that it's about 60+ mAP. I'm not sure......Sorry~

[developer0hye]

@yjh0410 Thanks~! I finished my M.S course. Your code helped me to graduate school.

[yjh0410]

Congratulations! I sincerely wish you more and more success in the future！ 어서!

------------------ 原始邮件 ------------------ 发件人: "yjh0410/yolov2-yolov3_PyTorch" <notifications@github.com>; 发送时间: 2020年9月13日(星期天) 晚上9:57 收件人: "yjh0410/yolov2-yolov3_PyTorch"<yolov2-yolov3_PyTorch@noreply.github.com>; 抄送: "ら ．Secret"<1394571815@qq.com>;"Mention"<mention@noreply.github.com>; 主题: Re: [yjh0410/yolov2-yolov3_PyTorch] mAP of yolov3-tiny on VOC (#38)

@yjh0410 Thanks~! I finished my M.S course. Your code helped me to graduate school.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[developer0hye]

@yjh0410 Thanks!! Haha, I really wish you more and more success in the future, too!!!

My yolov3 tiny model reached 53.51 mAP .

67.65% = aeroplane AP
70.54% = bicycle AP
47.78% = bird AP
37.30% = boat AP
24.25% = bottle AP
58.53% = bus AP
70.95% = car AP
61.54% = cat AP
29.46% = chair AP
55.86% = cow AP
43.19% = diningtable AP
57.06% = dog AP
66.56% = horse AP
68.29% = motorbike AP
63.59% = person AP
22.71% = pottedplant AP
55.82% = sheep AP
47.61% = sofa AP
65.59% = train AP
55.84% = tvmonitor AP
mAP = 53.51%

My model cannot detect pottedplant, chair, bottle, and boat well. Have you ever had this problem?

[yjh0410]

Thanks a lot ！！ I tried again on my devices, and got almost same result without multi-scale trick.

Maybe my tiny yolov3 is really too tiny to perform well.

The official tiny yolov3 doesn't provide its mAP on VOC, only the AP50 on COCO, so I don't have a reference to judge the quality of my model.

[developer0hye]

@yjh0410 Thanks a lot~!!

Does 'same result' mean that your model reached about 53.51 mAP?

Okay, it seems that the representation power of yolov3 tiny is too weak, so that model failed to detect some objects. When I found official yolov3 tiny's the mAP on VOC, I couldn't find this. So I asked you this. You are the one of the experts about detection algorithms that I know!

[yjh0410]

As far as I know，there‘s no information about tiny v3 on VOC.

It’s so weak that it isn't good at detecting many kinds of objects. Unless there is just one class like person or cat，tiny v3 can work well.

[developer0hye]

@yjh0410 Thanks!