def load_images(self,dataframe):
if dataframe is None:
return None
else:
assert type(dataframe) == pd.core.frame.DataFrame, "argument to load image should be dataframe"
assert "file_location" in dataframe.columns, "dataframe should contain file_location column"
output_images = np.zeros((len(dataframe),self.config.image_shape[0],self.config.image_shape[1],self.config.image_shape[2]))
for index,row in dataframe.iterrows():
img = cv2.imread(os.path.join(self.config.dataset_dir,row["file_location"].str_replace("mtk", "samuel")))
if img is None:
Log.WARNING("Unable to read images from "+os.path.join(self.config.dataset_dir,row["file_location"].str_replace("mtk", "samuel")))
continue
faces = self.detector(img)
for i in range(len(faces)):
if(len(faces)>0):
face_location = faces[i]
face_image = img[face_location.top():face_location.bottom(),face_location.left():face_location.right()]
try:
face_image = cv2.cvtColor(face_image,cv2.COLOR_BGR2GRAY)
face_image = cv2.resize(face_image,(self.config.image_shape[0],self.config.image_shape[1]))
output_images[index] = face_image.reshape(self.config.image_shape)
break
except:
# Log.ERROR_OUT = True
# Log.ERROR ("error"+","+str(face_image is None)+","+str(img is None)+","+ str(len(faces)))
# Log.ERROR (str(face_location.top())+","+ str(face_location.bottom())+","+str(face_location.left())+","+str(face_location.right()))
print (str(face_location.top())+","+ str(face_location.bottom())+","+str(face_location.left())+","+str(face_location.right()))
print ("error"+","+str(face_image is None)+","+str(img is None)+","+ str(len(faces)))
else:
face_image = cv2.cvtColor(face_image,cv2.COLOR_BGR2GRAY)
face_image = cv2.resize(img,(self.config.image_shape[0],self.config.image_shape[1]))
output_images[index] = face_image.reshape(self.config.image_shape)
Log.WARNING("Dlib unable to find faces from :"+os.path.join(self.config.dataset_dir,row["file_location"].str_replace("mtk", "samuel"))+" Loading full image as face")
return output_images
def load_dataset(self):
if self.config.label == "detection":
if not self.contain_dataset_files():
self.meet_convention()
Log.DEBUG_OUT = True
Log.DEBUG("Loading pickle files")
Log.DEBUG_OUT = False
self.train_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "train.pkl"))
self.test_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "test.pkl"))
if os.path.exists(
os.path.join(self.config.dataset_dir, "validation.pkl")):
self.validation_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "validation.pkl"))
else:
self.validation_dataset = None
frameinfo = getframeinfo(currentframe())
Log.WARNING("Unable to find validation dataset",
file_name=__name__,
line_number=frameinfo.lineno)
self.train_dataset = self.fix_labeling_issue(self.train_dataset)
self.test_dataset = self.fix_labeling_issue(self.test_dataset)
self.validation_dataset = self.fix_labeling_issue(
self.validation_dataset)
Log.DEBUG_OUT = True
Log.DEBUG("Loaded train, test and validation dataset")
Log.DEBUG_OUT = False
test_indexes = np.arange(len(self.test_dataset))
np.random.shuffle(test_indexes)
validation_indexes = np.arange(len(self.validation_dataset))
np.random.shuffle(validation_indexes)
self.test_dataset = self.test_dataset.iloc[
test_indexes].reset_index(drop=True)
self.validation_dataset = self.validation_dataset.iloc[
validation_indexes].reset_index(drop=True)
self.test_dataset = self.test_dataset[:1000]
self.validation_dataset = self.validation_dataset[:100]
Log.DEBUG_OUT = True
Log.DEBUG_OUT = False
self.test_dataset_images = self.load_images(
self.test_dataset).astype(np.float32) / 255
Log.DEBUG_OUT = True
Log.DEBUG("Loading validation images")
Log.DEBUG_OUT = False
self.validation_dataset_images = self.load_images(
self.validation_dataset).astype(np.float32) / 255
self.test_detection = self.test_dataset["is_face"].as_matrix()
self.dataset_loaded = True
Log.DEBUG_OUT = True
Log.DEBUG("Loaded all dataset and images")
Log.DEBUG_OUT = False
else:
raise NotImplementedError("Not implemented for labels:" +
str(self.labels))
def __init__(self,config):
self.config = config
self.model = AllInOneModel(self.config.image_shape)
if(config.model_weight!=None and os.path.exists(config.model_weight)):
Log.DEBUG_OUT = True
Log.DEBUG("Loading model weights from '"+config.model_weight+"'")
try:
self.model.model.load_weights(config.model_weight)
Log.DEBUG("Loaded model weights")
except:
Log.DEBUG("Unable to load model weight from "+config.model_weight)
Log.DEBUG_OUT =False
def load_images(self, dataframe):
if dataframe is None:
return None
else:
assert type(
dataframe
) == pd.core.frame.DataFrame, "argument to load image should be dataframe"
assert "file_location" in dataframe.columns, "dataframe should contain file_location column"
output_images = np.zeros(
(len(dataframe), self.config.image_shape[0],
self.config.image_shape[1], self.config.image_shape[2]))
for index, row in dataframe.iterrows():
img = cv2.imread(
os.path.join(self.config.dataset_dir,
row["file_location"][0]))
if img is None:
Log.WARNING("Unable to read images from " + os.path.join(
self.config.dataset_dir, row["file_location"][0]))
continue
face_location = row["face_location"][0].astype(int)
face_image = img[face_location[1]:face_location[3],
face_location[0]:face_location[2]]
face_image = cv2.cvtColor(face_image, cv2.COLOR_BGR2GRAY)
face_image = cv2.resize(
face_image,
(self.config.image_shape[0], self.config.image_shape[1]))
output_images[index] = face_image.reshape(
self.config.image_shape)
return output_images
def __init__(self,config):
self.config = config
self.model = AllInOneModel(self.config.image_shape)
self.model.save_model_to_json("/home/samuel/projects/All-In-One/AgeModel.json")
if(config.model_weight!=None and os.path.exists(config.model_weight)):
Log.DEBUG_OUT = True
Log.DEBUG("Loading model weights from '" + config.model_weight +"'")
try:
self.model.model.load_weights(config.model_weight)
#Freeze some layers in the network
"""
for layer in model.layers[:5]:
layers.trainable = False
"""
Log.DEBUG("Loaded model weights")
except:
Log.DEBUG("Unable to load model weight from "+config.model_weight)
Log.DEBUG_OUT =False
def load_dataset(self):
if set(self.labels).issubset(["Age", "Gender"]):
if not self.contain_dataset_files():
self.meet_convention()
Log.DEBUG_OUT = True
Log.DEBUG("Loading pickle files")
Log.DEBUG_OUT = False
self.train_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "train.pkl"))
self.test_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "test.pkl"))
if os.path.exists(
os.path.join(self.config.dataset_dir, "validation.pkl")):
self.validation_dataset = self.get_meta(
os.path.join(self.config.dataset_dir, "validation.pkl"))
else:
self.validation_dataset = None
frameinfo = getframeinfo(currentframe())
Log.WARINING_OUT = True
Log.WARNING("Unable to find validation dataset",
file_name=__name__,
line_number=frameinfo.lineno)
Log.WARINING_OUT = False
self.train_dataset = self.fix_labeling_issue(self.train_dataset)
self.test_dataset = self.fix_labeling_issue(self.test_dataset)
self.validation_dataset = self.fix_labeling_issue(
self.validation_dataset)
Log.DEBUG_OUT = True
Log.DEBUG("Loaded train, test and validation dataset")
Log.DEBUG_OUT = False
self.test_dataset = self.test_dataset[:5000]
self.validation_dataset = self.validation_dataset[:100]
Log.DEBUG_OUT = True
Log.DEBUG("Loading test images")
Log.DEBUG_OUT = False
self.test_dataset_images = self.load_images(
self.test_dataset).astype(np.float32) / 255
Log.DEBUG_OUT = True
Log.DEBUG("Loading validation images")
Log.DEBUG_OUT = False
self.validation_dataset_images = self.load_images(
self.validation_dataset).astype(np.float32) / 255
self.dataset_loaded = True
Log.DEBUG_OUT = True
Log.DEBUG("Loaded all dataset and images")
Log.DEBUG_OUT = False
else:
raise NotImplementedError("Not implemented for labels:" +
str(self.labels))