def main(unused_argv):
# Load data
train_x, train_y, val_x, val_y = load_training_validation_data(
train_view=train_view,
train_dir=train_dir,
val_view=val_view,
val_dir=val_dir)
train_data = np.asarray([flatten(x) for x in train_x], dtype=np.float32)
val_data = np.asarray([flatten(x) for x in val_x], dtype=np.float32)
train_labels = np.asarray(train_y, dtype=np.int32)
val_labels = np.asarray(val_y, dtype=np.int32)
# Create the Estimator
gei_classifier = tf.estimator.Estimator(model_fn=cnn_model_fn,
model_dir="./chpts")
# Set up logging for predictions
#tensors_to_log = {"probabilities": "softmax_tensor"}
#logging_hook = tf.train.LoggingTensorHook(
# tensors=tensors_to_log, every_n_iter=5000)
# Train the model
train_input_fn = tf.estimator.inputs.numpy_input_fn(x={"x": train_data},
y=train_labels,
batch_size=BATCH_SIZE,
num_epochs=None,
shuffle=True)
#gei_classifier.train(
# input_fn = train_input_fn,
# steps = 20000,
# hooks = [logging_hook])
gei_classifier.train(input_fn=train_input_fn, steps=20000)
# Evaluate the model and print results
eval_input_fn = tf.estimator.inputs.numpy_input_fn(x={"x": val_data},
y=val_labels,
num_epochs=1,
shuffle=False)
eval_results = gei_classifier.evaluate(input_fn=eval_input_fn)
print(eval_results)
def load_angle_train_data(img_class, view_list, train_dir):
human_id_list = ["%03d" % i for i in range(1, 51)]
human_id_list.remove('005')
human_id_list.remove('026') #126
human_id_list.remove('034')
human_id_list.remove('037') #144
human_id_list.remove('046')
human_id_list.remove('048') #54
if view_list is None:
view_list = [
"000", "018", "036", "054", "072", "090", "108", "126", "144",
"162", "180"
]
if train_dir is None:
train_dir = ["nm-%02d" % i for i in range(1, 5)]
training_x = []
training_y = []
# check dir exists
for id in human_id_list:
for dir in train_dir:
for view in view_list:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path,
id, dir, view)
if not os.path.exists(img_dir):
logger.error("%s do not exist" % img_dir)
for id in human_id_list:
logger.info("processing human %s" % id)
for dir in train_dir:
for view in view_list:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path,
id, dir, view)
data = img_path_to_IMG(img_class, img_dir)
if len(data.shape) > 0:
training_x.append(flatten(data))
training_y.append(view)
else:
print("LOAD_ANGLE_TRAIN_DATA: fail to extract %s of %s" %
(img_dir, id))
sys.exit(0)
return training_x, training_y
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
train_view = "090"
val_view = "090"
train_dir = ["nm-01"]
val_dir = ["nm-02"]
if __name__ == '__main__':
train_x, train_y, val_x, val_y = load_training_validation_data(
train_view=train_view,
train_dir=train_dir,
val_view=val_view,
val_dir=val_dir)
train_feature_x = [flatten(x) for x in train_x]
val_feature_x = [flatten(x) for x in val_x]
model = RandomForestClassification()
model.fit(x_train=train_feature_x, y_train=train_y)
predict_y = model.predict(val_feature_x)
print "predict_y: "
print predict_y
correct_count = sum(predict_y == val_y)
accuracy = correct_count * 1.0 / len(val_y)
print "train view: %s, val view: %s, accuracy: %d/%d=%.3f" % \
(train_view, val_view, correct_count, len(val_y), accuracy)
from data_tool import load_training_validation_data
from model.models import RandomForestClassification
from feature.hog import flatten
logger = logging.getLogger("main")
level = logging.INFO
log_filename = '%s.log' % __file__
format = '%(asctime)-12s[%(levelname)s] %(message)s'
datefmt ='%Y-%m-%d %H:%M:%S'
logging.basicConfig(level=level,
format=format,
filename=log_filename,
datefmt=datefmt)
if __name__ == '__main__':
train_dir = "train_dir"
val_dir = "test_dir"
training_x, training_y, validation_x, validation_y = load_training_validation_data(train_dir=train_dir, val_dir=val_dir)
training_feature_x = [flatten(x) for x in training_x]
validation_feature_x = [flatten(x) for x in validation_x]
logger.info("start")
model = RandomForestClassification()
model.fit(x_train=training_feature_x, y_train=training_y)
predict_file = "%s\model\predict.pickle" % config.Project.project_path
pickle.dump(model, open(predict_file, 'wb'))
predict_y = model.predict(validation_feature_x)
logger.info("done")
def load_data(img_class, data_class, probe_view, probe_dir, gallery_dir):
probe_type = probe_dir[0][0:2].upper()
if data_class == "validation":
human_id_list = ["%03d" % i for i in range(51, 75)]
human_id_list.remove('067')
human_id_list.remove('068')
elif data_class == "testing":
human_id_list = ["%03d" % i for i in range(75, 125)]
human_id_list.remove('079') #054
human_id_list.remove('088') #054
human_id_list.remove('109') #126
else:
print("Wrong data class")
sys.exit(0)
probe_x = []
probe_y = []
gallery_x = []
gallery_y = []
paired_x = []
paired_y = []
paired_data = []
# check dir exists
for id in human_id_list:
for dir in probe_dir:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path, id,
dir, probe_view)
if not os.path.exists(img_dir):
logger.error("%s do not exist" % img_dir)
for dir in gallery_dir:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path, id,
dir, probe_view)
if not os.path.exists(img_dir):
logger.error("%s do not exist" % img_dir)
# get probe list
for id in human_id_list:
logger.info("processing human %s" % id)
for dir in probe_dir:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path, id,
dir, probe_view)
probe_x.append(img_dir)
probe_y.append(id)
view_list = [
"000", "018", "036", "054", "072", "090", "108", "126", "144", "162",
"180"
]
# get gallery list
for id in human_id_list:
for dir in gallery_dir:
for view in view_list:
img_dir = "%s/%s/%s/%s" % (config.Project.casia_dataset_b_path,
id, dir, view)
gallery_x.append(img_dir)
gallery_y.append(id)
x_range = len(human_id_list) - 1
view_coe = len(view_list)
gallery_coe = len(view_list) * len(gallery_dir)
gallery_dir_range = len(gallery_dir) - 1
# get probe data
probe_imgs = [flatten(img_path_to_IMG(img_class, x)) for x in probe_x]
angles = get_angle(img_class, probe_type, np.asarray(probe_imgs))
#angles = np.asarray(len(probe_imgs)*[probe_view])
# test accuracy
#print(angles)
accuracy = sum(angles == [probe_view]) * 1.0 / len(angles)
print(accuracy)
for probe_idx, angle in enumerate(angles):
i = random.randint(0, gallery_dir_range)
v = view_list.index(angle)
gallery_idx = gallery_coe * (probe_idx //
len(probe_dir)) + view_coe * i + v
gallery_img = flatten(
img_path_to_IMG(img_class, gallery_x[gallery_idx]))
probe_img = probe_imgs[probe_idx]
if len(probe_img) > 0 and len(gallery_img) > 0:
paired_data.append(
[[np.asarray(probe_img),
np.asarray(gallery_img)], [0, 1]])
else:
print("LOAD_DATA: fail to extract %s of %s" % (img_dir, id))
x = random.randint(0, x_range)
gallery_idx = gallery_coe * (x) + view_coe * i + v
while (gallery_y[gallery_idx] == probe_y[probe_idx]):
x = random.randint(0, x_range)
gallery_idx = gallery_coe * (x) + view_coe * i + v
gallery_img = flatten(
img_path_to_IMG(img_class, gallery_x[gallery_idx]))
if len(probe_img) > 0 and len(gallery_img) > 0:
paired_data.append(
[[np.asarray(probe_img),
np.asarray(gallery_img)], [1, 0]])
else:
print("LOAD_DATA: fail to extract %s of %s" % (img_dir, id))
random.shuffle(paired_data)
paired_x = np.asarray([x.tolist() for x in np.asarray(paired_data)[:, 0]])
paired_y = np.asarray(paired_data)[:, 1]
return paired_x, paired_y
break
if event == 'Show Image':
window['-IMAGE-'].update(filename=values[0], size=(500, 200))
if event == 'Submit':
img_path = values[0]
values = []
img_dir = []
im = imread(img_path)
img_path = ''
img_dir.append(im)
data = img_path_to_GEI(img_dir)
img_dir = []
validation_x.append(data)
validation_feature_x = [flatten(x) for x in validation_x]
validation_x = []
predict_y = modell.predict(validation_feature_x)
validation_feature_x = []
age_group_ID = predict_y[0]
predict_y = []
age_group = ''
if age_group_ID == 'A':
age_group = '0 - 5'
elif age_group_ID == 'B':
age_group = '6 - 10'
elif age_group_ID == 'C':
age_group = '11 - 15'
elif age_group_ID == 'D':
