search

meituan / YOLOv6

YOLOv6: a single-stage object detection framework dedicated to industrial applications.
GNU General Public License v3.0
5.72k stars 1.04k forks source link

YOLOV6 在一些非常明显的前景大目标时无法检测，相比之下YOLOV5要很多好，希望能得到优化 #201

Closed tms2003 closed 2 years ago

tms2003 commented 2 years ago

https://www.youtube.com/watch?v=yd1DhOf8mW4的1:18秒 可以看出（同样的示例还有很多），YOLOV5要比YOLOV6好不少。YOLOV6对小目标检测不错。但对大目标有很大的问题。我认为这是应当不难得以改善的

mtjhl commented 2 years ago

Thank you for your attention, we will try to improve the mAP of large object.

mtjhl commented 2 years ago

I read your code, you user YOLOv5s6 trained with 1280 resolution to compare with YOLOv6s trained with 640 resolution, this is incomparable.

tms2003 commented 2 years ago

you should watch this:https://www.youtube.com/watch?v=5GXWvoDzpDU ,at 1 secomd clip ! you can found the truck be lost also!

lipond commented 2 years ago

@tms2003 兄弟，语法错误太多了，要么直接用中文吧，或者用机翻搞下都行啊

khwengXU commented 2 years ago

There are many factors can influence the detection effects for large/small objects. Could you give more details about the comparations? e.g. detailed model info/image resolution, etc.