Closed xianshunw closed 5 years ago
https://github.com/yun-liu/rcf/blob/cd007aca3bd6a908d9df2213fb62966606bb7bfa/examples/rcf/train_val.prototxt#L214
The feature map size of the conv_5 is 1/16 of the input image size, so I think the kernel size should be 32 and the stride should be 16.
Oh, sorry to disturb you. I did study your code carefully.
https://github.com/yun-liu/rcf/blob/cd007aca3bd6a908d9df2213fb62966606bb7bfa/examples/rcf/train_val.prototxt#L214
The feature map size of the conv_5 is 1/16 of the input image size, so I think the kernel size should be 32 and the stride should be 16.