caffe - Image as a label for pixel wise classification using lmdb -
i having problem configuring caffe pixel wise segmentation. however, after research, found can create imdb database labels. have done that, don't know how link data , label databases in training prototxt
file. example appreciated.
since aiming pixel-wise segmentation i'm gonna call call labels ground truths.
the simple solution - use 2 input layers.
use 1 input layer read image , fake labels. use input layer read ground truth , fake labels. use image , ground truth blobs wherever appropriate in network.
Comments
Post a Comment