Classifying with pretrained network - image resizing best practice
28 views
Skip to first unread message
Bhargava Manja
Jul 8, 2016, 2:52:45 PM7/8/16
to Caffe Users
If I am using a pretrained model (in my case vgg16) with a certain input size requirement (224x224) to classify arbitrary images, do I:
1. simply resize the image to the input size and ignore aspect ratio?
2. resize the image so the smallest side is the input size, and do some sort of cropping?
3. something else entirely?
Which would give the best classification results?
1. simply resize the image to the input size and ignore aspect ratio?
2. resize the image so the smallest side is the input size, and do some sort of cropping?
3. something else entirely?
Which would give the best classification results?
Reply all
Reply to author
Forward
0 new messages