How to train on different number of classes?

zhaoweicai / mscnn

Caffe implementation of our multi-scale object detection framework

405 stars 211 forks source link

How to train on different number of classes? #104

Closed liz2020 closed 5 years ago

liz2020 commented 5 years ago

I want to use the cascade mscnn on a new dataset (in kitti formate), which have two classes 'car' and 'person'. So I modify the cls_name in trainval_1st.prototxt, trainval_2nd.prototxt, and mscnn_deploy.prototxt to 3 (2 class + 1 background), and change all the "num_output : 9 " to "num_output: 7". Is this the correct way to follow?

CassieMai commented 5 years ago

@liz2020 Should the background be added to the number of classes?

liz2020 commented 5 years ago

Yes, the number of classes includes the background.

CassieMai commented 5 years ago

Thank you @liz2020