meituan / YOLOv6

YOLOv6: a single-stage object detection framework dedicated to industrial applications.
GNU General Public License v3.0
5.7k stars 1.03k forks source link

Model training and evaluation problems #277

Closed Guan-LinHe closed 2 years ago

Guan-LinHe commented 2 years ago

This is the model after we train the model, but I want to get the P(precision), R(recall), F1score of the model how to get these values. image

mtjhl commented 2 years ago

Hi, if you want to get precision and recall at one specific confidence, you can filter out the detect boxes whose confidence are lower than your specific confidence. Second, put the detect results and annotation to pycocotools, you can get the details contain precision and recall at cocoEval here. https://github.com/meituan/YOLOv6/blob/c85065cf4b58c7f929013fbc73c8dc362e8bed86/yolov6/core/evaler.py#L135 cocoEval contains rich information. You can explore it.

Guan-LinHe commented 2 years ago

@mtjhl Sorry, my problem is to other detection frameworks (ex: yolov5, yolor, yolox), when the training is completed, precision and recall can be found in .txt or .csv, but I only have .json in yolov6 and do not see the above precision and recall recall, can you please assist me? Thank you.

mtjhl commented 2 years ago

Ok, we will discuss about whether to add this function.

Guan-LinHe commented 2 years ago

@mtjhl So I can't know about precision and recall?There are other ways...

HeZhang33 commented 2 years ago

I have just asked the similar feature request. Hope to have precision and recall output as well. #266

Passion0930 commented 2 years ago

Ok, we will discuss about whether to add this function.

yes, we want to know precision and recall of all kinds of categories. please add this function soon!

Chilicyy commented 2 years ago

Thanks for your suggestion! We'll mark it down and try to add this feature.

norah-art commented 2 years ago

Hi, Is there any update on that? It will be amazing for having (precision), (recall), F1 score

shensheng272 commented 2 years ago

Latest code will print conf on best mF1 and print P R under iou50. Also default plot curves like yolov5. Also you can log infos you want near https://github.com/meituan/YOLOv6/blob/main/yolov6/core/evaler.py#L188