Skip to first unread message
Niko Gamulin
May 29, 2015, 4:41:29 PM5/29/15
to caffe...@googlegroups.com
Hi,
In order to use lmdb or leveldb database instead of hdf5 to classify the data with multiple labels, is it OK to slice the labels?
For example, if there are 5 binary outputs, is it possible to prepare the lmdb dataset with images and labels as follows and then slice the label:
img1.jpeg 10101
img2.jpeg 11100
....
If the data layer is defined as follows:
layers {
name: "mnist"
type: DATA
top: "data"
top: "label"
data_param {
source: "/home/niko/datasets/train_lmdb"
backend: LMDB
batch_size: 32
}
transform_param {
crop_size: 224
mean_file: "/home/niko/datasets/_mean.binaryproto"
mirror: true
}
include: { phase: TRAIN }
}is it possible (convenient) to slice the labels the following way:
layers {
name: "slicer_label"
type: SLICE
bottom: "label"
top: "label1"
top: "label2"
top: "label3"
top: "label4"
top: "label5"
slice_param {
slice_dim: 1
slice_point: 1
slice_point: 2
slice_point: 3
slice_point: 4
}
}
Anyway, I would be thankful for any comments or suggested modifications, necessary to split the 5-digit label into 5 single-digit labels.
Best regards,
Niko
Reply all
Reply to author
Forward
0 new messages