Skip to first unread message
Cuong Duc
Aug 14, 2015, 3:42:26 PM8/14/15
to Caffe Users
Hi,
I want to train CaffeNet with my own dataset. 100k images for training and 26k for validation.
I don't quite get the meaning of 'test_iter' in solver.prototxt file
The default is test_iter : 1000. However, my current training batch size is 96, so is this correct that I can set the test_iter to around 270 as 96*270 ~ 26K?
I want to train CaffeNet with my own dataset. 100k images for training and 26k for validation.
I don't quite get the meaning of 'test_iter' in solver.prototxt file
The default is test_iter : 1000. However, my current training batch size is 96, so is this correct that I can set the test_iter to around 270 as 96*270 ~ 26K?
Oscar Beijbom
Aug 15, 2015, 1:52:18 PM8/15/15
to Caffe Users
It means the number of batches to run on the test set. So yeah, if you want the test phase to run on your whole val. set, then yeah, 270 is a good value. But you don't need to run on the whole test-set every time.
cheers
Reply all
Reply to author
Forward
0 new messages