Skip to first unread message
elliot...@gmail.com
May 12, 2016, 3:19:57 AM5/12/16
to Caffe Users
I'm using the model trained by cifar10 data set to classify images I collect on the Internet, but the accuracy is poor low.
I guess the reason is because training images in cifar10 have a property that a large proportion of the total area is the object instance while images collected on the Internet don't have this property.
Am I guessing right? Has anyone used the cifar10 model on his own images having the same problem like me?
I guess the reason is because training images in cifar10 have a property that a large proportion of the total area is the object instance while images collected on the Internet don't have this property.
Am I guessing right? Has anyone used the cifar10 model on his own images having the same problem like me?
Reply all
Reply to author
Forward
0 new messages