def test_rpn_hog():
"""
HOG+RPN for panel label recognition
:return:
"""
parser = OptionParser()
parser.add_option("-p", "--path", dest="test_path", help="Path to test data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option("-n", "--num_rois", type="int", dest="num_rois",
help="Number of ROIs per iteration. Higher means more memory use.", default=32)
parser.add_option("--config_filename", dest="config_filename",
help="Location to read the metadata related to the training (generated when training).",
default="config.pickle")
parser.add_option("--network", dest="network", help="Base network to use. Supports nn_cnn_3_layer.",
default='nn_cnn_3_layer')
parser.add_option("--rpn_weight_path", dest="rpn_weight_path", default='./model_rpn.hdf5')
# default='/Users/jie/projects/PanelSeg/ExpPython/models/label+bg_rpn_3_layer_color-0.0374.hdf5')
parser.add_option("--classify_model_path", dest="classify_model_path",
default='/Users/jie/projects/PanelSeg/ExpPython/models/label50+bg_cnn_3_layer_color-0.9910.h5')
parser.add_option("--result_folder", dest="result_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn_cpu_0.116')
(options, args) = parser.parse_args()
if not options.test_path: # if filename is not given
parser.error('Error: path to test data must be specified. Pass --path to command line')
rpn_c, rpn_model_rpn, rpn_model_classifier, hog = rpn_hog_initialize(options)
with open(options.test_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if 'PMC3664797_gkt198f2p' not in filepath:
# continue
# if idx < 243:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
st = time.time()
figure.fg_rois, figure.fg_scores, figure.fg_labels = rpn_hog_detect(figure,
rpn_c, rpn_model_rpn, rpn_model_classifier,
hog, RPN_ONLY)
print('Elapsed time = {}'.format(time.time() - st))
# Save detection results
figure.save_annotation(options.result_folder)
def test_lstm():
parser = OptionParser()
parser.add_option("--eval_path", dest="eval_path", help="Path to eval data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option("--eval_auto_folder", dest="eval_auto_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn')
parser.add_option("--classify_model_path", dest="classify_model_path",
default='/Users/jie/projects/PanelSeg/ExpPython/models/label50+bg_cnn_3_layer_color-0.9910.h5')
# parser.add_option("--svm_model_path", dest="svm_model_path",
# default='/Users/jie/projects/PanelSeg/Exp/LabelClassifySvmTrain/SVMModel-51classes-with-neg/svm_model_rbf_8.0_0.125')
parser.add_option("--lstm_model_path", dest="lstm_model_path",
default='/Users/jie/projects/PanelSeg/ExpPython/models/lstm_model_train_0.25eval_epoch_9.h5')
parser.add_option("--result_folder", dest="result_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog_lstm/eval')
(options, args) = parser.parse_args()
# model_classifier = load_model(options.classify_model_path)
# model_classifier.summary()
# model_svm = svmutil.svm_load_model(options.svm_model_path)
model_svm = None
model_lstm = load_model(options.lstm_model_path)
model_lstm.summary()
with open(options.eval_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if '1757-1626-0002-0000008402-001' not in filepath:
# continue
# if idx < 37:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
st = time.time()
# load detection results by RPN
rois = load_auto_annotation(figure, options.eval_auto_folder)
# sort auto annotation with respect to distances to left-up corner (0, 0)
distances = [roi[0] + roi[1] for roi in rois]
indexes = np.argsort(distances)
rois = rois[indexes]
x = feature_extraction(figure, rois)
if rois.size == 0:
figure.fg_rois, figure.fg_scores, figure.fg_labels = None, None, None
else:
# x = x.tolist()
# y = np.zeros(len(x)).tolist()
# p_label, p_acc, p_val = svmutil.svm_predict(y, x, model_svm, '-b 1')
# x = np.array(p_val)
_x = np.expand_dims(x, axis=0)
y_hat = model_lstm.predict(_x)
# figure.fg_rois, figure.fg_scores, figure.fg_labels = max_y_hat(rois, y_hat[0])
figure.fg_rois, figure.fg_scores, figure.fg_labels = beam_search_with_neg(rois, y_hat[0], 5)
print('Elapsed time = {}'.format(time.time() - st))
# Save detection results
figure.save_annotation(options.result_folder)
def combine_hog_rpn():
"""
Load results from HOG and RPN methods and then combine them
:return:
"""
parser = OptionParser()
parser.add_option("-p", "--path", dest="test_path", help="Path to test data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option("--hog_folder", dest="hog_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/hog')
parser.add_option("--rpn_folder", dest="rpn_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn_cpu_0.0374')
parser.add_option("--result_folder", dest="result_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/hog_rpn')
(options, args) = parser.parse_args()
with open(options.test_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if 'PMC3664797_gkt198f2p' not in filepath:
# continue
# if idx < 243:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
# Load HOG result
hog_file_path = os.path.join(options.hog_folder, figure.file).replace('.jpg', '_data.xml')
figure.load_annotation(hog_file_path)
hog_rois = np.empty([len(figure.panels), 4], dtype=int)
for i in range(len(figure.panels)):
hog_rois[i] = figure.panels[i].label_rect
hog_rois[:, 0] += Figure.PADDING
hog_rois[:, 1] += Figure.PADDING
hog_labels = np.full(hog_rois.shape[0], Panel.LABEL_ALL)
hog_scores = np.full(hog_rois.shape[0], 1.0)
# Load RPN result
rpn_file_path = os.path.join(options.rpn_folder, figure.file).replace('.jpg', '_data.xml')
figure.load_annotation(rpn_file_path)
rpn_rois = np.empty([len(figure.panels), 4], dtype=int)
for i in range(len(figure.panels)):
rpn_rois[i] = figure.panels[i].label_rect
rpn_rois[:, 0] += Figure.PADDING
rpn_rois[:, 1] += Figure.PADDING
rpn_labels = np.full(rpn_rois.shape[0], Panel.LABEL_ALL)
rpn_scores = np.full(rpn_rois.shape[0], 1.0)
# Keep all regions detected by both RPN and HOG methods.
# rois, scores, labels = rpn_hog_combine_all(hog_rois, hog_scores, hog_labels, rpn_rois, rpn_scores, rpn_labels)
# Keep only regions agreed by both methods. IOU > 25%
rois, scores, labels = rpn_hog_combine_agreed(hog_rois, hog_scores, hog_labels,
rpn_rois, rpn_scores, rpn_labels, 0.125)
figure.fg_rois, figure.fg_scores, figure.fg_labels = rois, scores, labels
# Save detection results
figure.save_annotation(options.result_folder)
def test_lstm():
parser = OptionParser()
parser.add_option(
"--eval_path",
dest="eval_path",
help="Path to eval data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option(
"--eval_auto_folder",
dest="eval_auto_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn')
parser.add_option(
"--classify_model_path",
dest="classify_model_path",
default=
'/Users/jie/projects/PanelSeg/ExpPython/models/label50+bg_cnn_3_layer_color-0.9910.h5'
)
# parser.add_option("--svm_model_path", dest="svm_model_path",
# default='/Users/jie/projects/PanelSeg/Exp/LabelClassifySvmTrain/SVMModel-51classes-with-neg/svm_model_rbf_8.0_0.125')
parser.add_option(
"--lstm_model_path",
dest="lstm_model_path",
default=
'/Users/jie/projects/PanelSeg/ExpPython/models/lstm_model_train_0.25eval_epoch_9.h5'
)
parser.add_option(
"--result_folder",
dest="result_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog_lstm/eval'
)
(options, args) = parser.parse_args()
# model_classifier = load_model(options.classify_model_path)
# model_classifier.summary()
# model_svm = svmutil.svm_load_model(options.svm_model_path)
model_svm = None
model_lstm = load_model(options.lstm_model_path)
model_lstm.summary()
with open(options.eval_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if '1757-1626-0002-0000008402-001' not in filepath:
# continue
# if idx < 37:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
st = time.time()
# load detection results by RPN
rois = load_auto_annotation(figure, options.eval_auto_folder)
# sort auto annotation with respect to distances to left-up corner (0, 0)
distances = [roi[0] + roi[1] for roi in rois]
indexes = np.argsort(distances)
rois = rois[indexes]
x = feature_extraction(figure, rois)
if rois.size == 0:
figure.fg_rois, figure.fg_scores, figure.fg_labels = None, None, None
else:
# x = x.tolist()
# y = np.zeros(len(x)).tolist()
# p_label, p_acc, p_val = svmutil.svm_predict(y, x, model_svm, '-b 1')
# x = np.array(p_val)
_x = np.expand_dims(x, axis=0)
y_hat = model_lstm.predict(_x)
# figure.fg_rois, figure.fg_scores, figure.fg_labels = max_y_hat(rois, y_hat[0])
figure.fg_rois, figure.fg_scores, figure.fg_labels = beam_search_with_neg(
rois, y_hat[0], 5)
print('Elapsed time = {}'.format(time.time() - st))
# Save detection results
figure.save_annotation(options.result_folder)
def combine_hog_rpn():
"""
Load results from HOG and RPN methods and then combine them
:return:
"""
parser = OptionParser()
parser.add_option(
"-p",
"--path",
dest="test_path",
help="Path to test data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option(
"--hog_folder",
dest="hog_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/hog')
parser.add_option(
"--rpn_folder",
dest="rpn_folder",
default=
'/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn_cpu_0.0374')
parser.add_option(
"--result_folder",
dest="result_folder",
default='/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/hog_rpn')
(options, args) = parser.parse_args()
with open(options.test_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if 'PMC3664797_gkt198f2p' not in filepath:
# continue
# if idx < 243:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
# Load HOG result
hog_file_path = os.path.join(options.hog_folder,
figure.file).replace('.jpg', '_data.xml')
figure.load_annotation(hog_file_path)
hog_rois = np.empty([len(figure.panels), 4], dtype=int)
for i in range(len(figure.panels)):
hog_rois[i] = figure.panels[i].label_rect
hog_rois[:, 0] += Figure.PADDING
hog_rois[:, 1] += Figure.PADDING
hog_labels = np.full(hog_rois.shape[0], Panel.LABEL_ALL)
hog_scores = np.full(hog_rois.shape[0], 1.0)
# Load RPN result
rpn_file_path = os.path.join(options.rpn_folder,
figure.file).replace('.jpg', '_data.xml')
figure.load_annotation(rpn_file_path)
rpn_rois = np.empty([len(figure.panels), 4], dtype=int)
for i in range(len(figure.panels)):
rpn_rois[i] = figure.panels[i].label_rect
rpn_rois[:, 0] += Figure.PADDING
rpn_rois[:, 1] += Figure.PADDING
rpn_labels = np.full(rpn_rois.shape[0], Panel.LABEL_ALL)
rpn_scores = np.full(rpn_rois.shape[0], 1.0)
# Keep all regions detected by both RPN and HOG methods.
# rois, scores, labels = rpn_hog_combine_all(hog_rois, hog_scores, hog_labels, rpn_rois, rpn_scores, rpn_labels)
# Keep only regions agreed by both methods. IOU > 25%
rois, scores, labels = rpn_hog_combine_agreed(hog_rois, hog_scores,
hog_labels, rpn_rois,
rpn_scores, rpn_labels,
0.125)
figure.fg_rois, figure.fg_scores, figure.fg_labels = rois, scores, labels
# Save detection results
figure.save_annotation(options.result_folder)
def test_rpn_hog():
"""
HOG+RPN for panel label recognition
:return:
"""
parser = OptionParser()
parser.add_option(
"-p",
"--path",
dest="test_path",
help="Path to test data.",
default='/Users/jie/projects/PanelSeg/ExpPython/eval.txt')
parser.add_option(
"-n",
"--num_rois",
type="int",
dest="num_rois",
help="Number of ROIs per iteration. Higher means more memory use.",
default=32)
parser.add_option(
"--config_filename",
dest="config_filename",
help=
"Location to read the metadata related to the training (generated when training).",
default="config.pickle")
parser.add_option("--network",
dest="network",
help="Base network to use. Supports nn_cnn_3_layer.",
default='nn_cnn_3_layer')
parser.add_option("--rpn_weight_path",
dest="rpn_weight_path",
default='./model_rpn.hdf5')
# default='/Users/jie/projects/PanelSeg/ExpPython/models/label+bg_rpn_3_layer_color-0.0374.hdf5')
parser.add_option(
"--classify_model_path",
dest="classify_model_path",
default=
'/Users/jie/projects/PanelSeg/ExpPython/models/label50+bg_cnn_3_layer_color-0.9910.h5'
)
parser.add_option(
"--result_folder",
dest="result_folder",
default=
'/Users/jie/projects/PanelSeg/ExpPython/eval/rpn_hog/rpn_cpu_0.116')
(options, args) = parser.parse_args()
if not options.test_path: # if filename is not given
parser.error(
'Error: path to test data must be specified. Pass --path to command line'
)
rpn_c, rpn_model_rpn, rpn_model_classifier, hog = rpn_hog_initialize(
options)
with open(options.test_path) as f:
lines = f.readlines()
for idx, filepath in enumerate(lines):
print(str(idx) + ': ' + filepath)
# if 'PMC3664797_gkt198f2p' not in filepath:
# continue
# if idx < 243:
# continue
filepath = filepath.strip()
figure = Figure(filepath)
figure.load_image()
st = time.time()
figure.fg_rois, figure.fg_scores, figure.fg_labels = rpn_hog_detect(
figure, rpn_c, rpn_model_rpn, rpn_model_classifier, hog, RPN_ONLY)
print('Elapsed time = {}'.format(time.time() - st))
# Save detection results
figure.save_annotation(options.result_folder)
