-
Run matlab script: create_labels.m . [OUTPUT = train_list.txt or test_list.txt] This script creates a file train_list.txt that associates to each image its label. It assumes that images are saved in a folder named as the object label. The script also converts the images from .ppm (output of yarpdatadumper) into .jpg.
-
Open bash script create_database.sh, change path in DATA and run it [OUTPUT = train_database or test_database] [OUTPUT = train_mean.binaryproto or test_mean.binaryproto] This script creates a database starting from the images and the train_list.txt file and the mean images.
- From dev/object_reco, run
./../caffe/build/tools/caffe train --solver=models/icub_alexnet/solver.prototxt