训练时如何加入无标注框的负样本来减少模型误判率 #702
-
|
在自定义数据集中训练模型进行测试，总会发现部分无目标的图片出现误判的现象，mmyolo能否支持加入这些负样本到训练集中进行训练？就像PaddleDetection中支持加载无标注框进行训练。 |
Beta Was this translation helpful? Give feedback.
Answered by
hhaAndroid
Apr 17, 2023
Replies: 1 comment 1 reply
-
|
已经支持了。https://mmyolo.readthedocs.io/zh_CN/dev/recommended_topics/training_testing_tricks.html#id6 |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
好的谢谢，那这些纯背景图片是要如何加入到自定义的coco数据集中？就在标注的时候要怎么操作？或者在coco的json中是怎样的格式？ |
Beta Was this translation helpful? Give feedback.
Answer selected by
hhaAndroid
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
已经支持了。https://mmyolo.readthedocs.io/zh_CN/dev/recommended_topics/training_testing_tricks.html#id6