为什么Precision和Recall一直处在一个奇怪的值？
diplomatist opened this issue · comments
from 0 - 21 epoch ：precision=recall =0
from 22- 199 epoch: precision = 1 recall = 0.000231
我跑的是自己的数据集。
同等超参下能达到比YOLOV7还高的MAP，这令人非常疑惑。
你好 你解决这个问题了吗
你好 你解决这个问题了吗
没有解决
It seems to related with cfg file.
if you train by custom dataset, change filter and class in cfg file
你好 你解决这个问题了吗
#413 (comment)
如果用自定义数据集，cfg文件中有两个地方要修改：
1-yolo上一层的卷积filter修改成3*(classes+5)
2-yolo层的classes=[classnum of your datasets]
It seems to related with cfg file. if you train by custom dataset, change filter and class in cfg file
Thank you for your answer. I have solved this problem through your proposal. It is normal now.
