g_part_image_semSeg_embedding_testing_folder = '/media/adrian/Datasets/datasets/image_embedding/combinedShape_part_image_semSeg_embedding_testing_03001627_manifoldNet'
# END BORRAR - solo para manifold combinado
# Loop this for every part
part_id = 4
g_network_architecture_name = 'manifoldNet'
g_part_image_embedding_training_prototxt = os.path.join(g_part_image_semSeg_embedding_training_folder, 'image_embedding_'+g_network_architecture_name+'_part'+str(part_id)+'.prototxt')
g_part_image_embedding_testing_prototxt = os.path.join(g_part_image_semSeg_embedding_testing_folder, 'image_embedding_'+g_network_architecture_name+'_part'+str(part_id)+'.prototxt')
image_embedding_testing_in = os.path.join(BASE_DIR, 'image_embedding_'+g_network_architecture_name+'.prototxt.in')
print 'Preparing %s...' % g_part_image_embedding_testing_prototxt
shutil.copy(image_embedding_testing_in, g_part_image_embedding_testing_prototxt)
for line in fileinput.input(g_part_image_embedding_testing_prototxt, inplace=True):
line = line.replace('embedding_space_dim', str(g_shape_embedding_space_dimension))
line = line.replace('partX', 'part'+str(part_id))
sys.stdout.write(line)
part_image_semSeg_embedding_caffemodel = os.path.join(g_part_image_semSeg_embedding_training_folder, 'single_manifold_snapshots', 'snapshots%s_part%d_iter_%d.caffemodel' % (g_shapenet_synset_set_handle, part_id, args.iter_num))
part_image_semSeg_embedding_caffemodel_stacked = os.path.join(g_part_image_semSeg_embedding_testing_folder, 'stacked%s_part%d_iter_%d.caffemodel' % (g_shapenet_synset_set_handle, part_id, args.iter_num))
stack_caffe_models(prototxt=g_part_image_embedding_testing_prototxt,
base_model=g_extract_feat_pool5_caffemodel,
top_model=part_image_semSeg_embedding_caffemodel,
stacked_model=part_image_semSeg_embedding_caffemodel_stacked,
caffe_path=g_caffe_install_path)
# -*- coding: utf-8 -*-
import os, sys
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.dirname(BASE_DIR))
from global_variables import *
from utilities_caffe import extract_cnn_features, stack_caffe_models
# Contains the feature layer training data
# g_extract_feat_manifold_prototxt = '/media/adrian/Datasets/datasets/image_embedding/part_image_semSeg_embedding_training_03001627_rcnn/train_val_rcnn.prototxt'
extract_feat_manifold_caffemodel = '/media/adrian/Datasets/datasets/image_embedding/part_image_semSeg_embedding_training_03001627_rcnn/snapshots/snapshots_03001627_iter_100.caffemodel'
stack_caffe_models(prototxt=g_extract_feat_pool5_prototxt,
base_model=g_fine_tune_manifold_caffemodel,
top_model=extract_feat_manifold_caffemodel,
stacked_model=g_extract_feat_pool5_caffemodel,
caffe_path=g_caffe_install_path)
extract_cnn_features(img_filelist=g_syn_images_filelist,
img_root='/',
prototxt=g_extract_feat_pool5_prototxt,
caffemodel=g_extract_feat_pool5_caffemodel,
feat_name='manifold_pool5',
output_path=g_pool5_semSeg_lmdb,
output_type='lmdb',
caffe_path=g_caffe_install_path,
mean_file=g_mean_file,
gpu_index=g_extract_feat_gpu_index,
pool_size=g_extract_feat_thread_num)
parser.add_argument(
'--iter_num',
'-n',
help='Use image embedding model trained after iter_num iterations',
type=int,
default=40000)
args = parser.parse_args()
image_embedding_testing_in = os.path.join(
BASE_DIR,
'image_embedding_' + g_network_architecture_name + '.prototxt.in')
print 'Preparing %s...' % (g_image_embedding_testing_prototxt)
shutil.copy(image_embedding_testing_in, g_image_embedding_testing_prototxt)
for line in fileinput.input(g_image_embedding_testing_prototxt, inplace=True):
line = line.replace('embedding_space_dim',
str(g_shape_embedding_space_dimension))
sys.stdout.write(line)
image_embedding_caffemodel = os.path.join(
g_image_embedding_training_folder, 'snapshots',
'snapshots%s_iter_%d.caffemodel' %
(g_shapenet_synset_set_handle, args.iter_num))
image_embedding_caffemodel_stacked = os.path.join(
g_image_embedding_testing_folder, 'snapshots%s_iter_%d.caffemodel' %
(g_shapenet_synset_set_handle, args.iter_num))
stack_caffe_models(prototxt=g_image_embedding_testing_prototxt,
base_model=g_fine_tune_caffemodel,
top_model=image_embedding_caffemodel,
stacked_model=image_embedding_caffemodel_stacked,
caffe_path=g_caffe_install_path)
import argparse
import fileinput
#https://github.com/BVLC/caffe/issues/861#issuecomment-70124809
import matplotlib
matplotlib.use('Agg')
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.dirname(BASE_DIR))
from global_variables import *
from utilities_caffe import stack_caffe_models
parser = argparse.ArgumentParser(description="Stitch pool5 extraction and image embedding caffemodels together.")
parser.add_argument('--iter_num', '-n', help='Use image embedding model trained after iter_num iterations', type=int, default=20000)
args = parser.parse_args()
image_embedding_testing_in = os.path.join(BASE_DIR, 'image_embedding_'+g_network_architecture_name+'.prototxt.in')
print 'Preparing %s...'%(g_image_embedding_testing_prototxt)
shutil.copy(image_embedding_testing_in, g_image_embedding_testing_prototxt)
for line in fileinput.input(g_image_embedding_testing_prototxt, inplace=True):
line = line.replace('embedding_space_dim', str(g_shape_embedding_space_dimension))
sys.stdout.write(line)
image_embedding_caffemodel = os.path.join(g_image_embedding_training_folder, 'snapshots%s_iter_%d.caffemodel'%(g_shapenet_synset_set_handle, args.iter_num))
image_embedding_caffemodel_stacked = os.path.join(g_image_embedding_testing_folder, 'snapshots%s_iter_%d.caffemodel'%(g_shapenet_synset_set_handle, args.iter_num))
stack_caffe_models(prototxt=g_image_embedding_testing_prototxt,
base_model=g_fine_tune_caffemodel,
top_model=image_embedding_caffemodel,
stacked_model=image_embedding_caffemodel_stacked,
caffe_path=g_caffe_install_path)