def __init__(self, folderpath):
"""Constructor."""
self.__trainingpath__ = None
self.__evaluatepath__ = None
self.__expressions__ = []
self.__trainingset__ = {}
self.__evaluateset__ = {}
subfolders = os.listdir(folderpath)
# Check for trailing '/'
if folderpath[-1:] != '/':
folderpath += '/'
self.__trainingpath__ = folderpath
# If we got the training & evaluate subfolders
if "training" in subfolders:
self.__trainingpath__ = folderpath+"training/"
if "evaluate" in subfolders:
self.__evaluatepath__ = folderpath+"evaluate/"
# get the classes
exps = os.listdir(self.__trainingpath__)
for i, e in enumerate(exps):
if Exp.from_str(e) is None:
print("Unknown expression: "+e)
else:
self.__expressions__.append(Exp.from_str(e))
# Init the sets
for i, e in enumerate(self.__expressions__):
self.__trainingset__[e] = []
self.__evaluateset__[e] = []
def load_videos(self):
"""Load pictures from the tree structure."""
exps = os.listdir(self.__trainingpath__)
for i, e in enumerate(exps):
if Exp.from_str(e) is None:
print("Unknown expression: "+e)
else:
exp = Exp.from_str(e)
path = self.__trainingpath__+e+'/'
for id, f in enumerate(glob.glob(os.path.join(path, "*.avi"))):
pics = []
cap = cv2.VideoCapture(f, cv2.CAP_FFMPEG)
while(cap.isOpened()):
ret, frame = cap.read()
if ret:
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
pics.append(frame)
else:
cap.release()
self.__trainingset__[exp].append(pics)
if self.__evaluatepath__ is not None:
exps = os.listdir(self.__evaluatepath__)
for i, e in enumerate(exps):
if Exp.from_str(e) is None:
print("Unknown expression: "+e)
else:
exp = Exp.from_str(e)
path = self.__evaluatepath__+e+'/'
for id, f in enumerate(glob.glob(os.path.join(path,
"*.avi"))):
pics = []
cap = cv2.VideoCapture(f, cv2.CAP_FFMPEG)
while(cap.isOpened()):
ret, frame = cap.read()
if ret:
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
pics.append(frame)
else:
cap.release()
self.__trainingset__[exp].append(pics)
def load_pictures(self):
"""Load pictures from the tree structure."""
exps = os.listdir(self.__trainingpath__)
for i, e in enumerate(exps):
print(e)
if Exp.from_str(e) is None:
print("Unknown expression: "+e)
else:
exp = Exp.from_str(e)
path = self.__trainingpath__+e+'/'
source_map = open(path+"map.csv")
line = source_map.readline() # First line, headers, ignore.
line = source_map.readline()
while line != '':
line = line.split(',')
pic = cv2.imread(line[1].replace('\n', ''))
pic = cv2.cvtColor(pic, cv2.COLOR_BGR2GRAY)
self.__trainingset__[exp].append((pic, line[0]))
line = source_map.readline()
if self.__evaluatepath__ is not None:
exps = os.listdir(self.__evaluatepath__)
for i, e in enumerate(exps):
if Exp.from_str(e) is None:
print("Unknown expression: "+e)
else:
exp = Exp.from_str(e)
path = self.__evaluatepath__+e+'/'
for id, f in enumerate(glob.glob(os.path.join(path,
"*.jpg"))):
source_map = open(path+"map.csv")
line = source_map.readline()
line = source_map.readline()
while line != '':
line = line.split(',')
pic = cv2.imread(line[1].replace('\n', ''))
pic = cv2.cvtColor(pic, cv2.COLOR_BGR2GRAY)
self.__evaluateset__[exp].append((pic, line[0]))
line = source_map.readline()
def get_dataset_from_multiple_json(folder_path):
datas = FlandmarksDataStorage()
for id, f in enumerate(glob.glob(os.path.join(folder_path, "*.json"))):
# Open the data file
src = json.load(open(f, 'r'))["datas"]
for j, src_elem in enumerate(src):
data = FlandmarksData(src_elem["file"],
Exp.from_str(src_elem["oclass"]))
# Save the n frames points
for pos in range(0, len(src_elem["values"])):
if src_elem["values"][pos] is not None:
fl = FLandmarks()
fl.set_points(src_elem["values"][pos]["points"])
data.add_flandmarks(fl)
else:
data.add_flandmarks(None)
datas.add_element(data)
return datas
model = load_model(model_path)
labels = [
Exp.ANGER, Exp.DISGUST, Exp.FEAR, Exp.HAPPINESS, Exp.SADNESS, Exp.SURPRISE,
Exp.NEUTRAL
]
# Write the header of the output csv file.
output_file = open(output_path, 'w')
output_file.write('filepath,' + (',').join([l.to_str()
for l in labels]) + '\n')
# For each extracted landmarks csv file, we do the prediction.
for f in open_landmarks_files(source_path):
for filepath, exp, datas in group_line_by_file(f):
exp = Exp.from_str(exp[0])
# Retrieve the faces frame by frame
cap = cv2.VideoCapture(filepath, cv2.CAP_FFMPEG)
frameid = 0
frames = []
while (cap.isOpened()):
ret, frame = cap.read()
if ret:
filepath, frame_id, exp, bounds, points = datas[frameid]
# The dlib algorithm might not succeeded in finding a face.
# In such case, the bounding box is equal to None. So we'll do
# the face extraction on the frame which actually have a
# bounding box.
if bounds is not None:
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
frame = extract_face_frame(frame, bounds, (64, 64), 15)
line = cnn_file.readline()
while line != '':
line = line.split(',')
file = line[0]
file = file.split('/')[-1]
values = [float(v) * 100 for v in line[1:]]
values = ["%.2f" % round(v, 2) for v in values]
file_values_map[file] = [values]
line = cnn_file.readline()
# Parse the reviewers labels
line = reviewers_file.readline()
reviewers_labels = line.split(',')[24:30]
reviewers_labels = [s.replace('(%)', '') for s in reviewers_labels]
reviewers_labels = [Exp.from_str(s) for s in reviewers_labels]
# We now add the NEUTRAL label to replace the "Undecided" label, for more
# consistance with the cnn labels.
reviewers_labels.append(Exp.NEUTRAL)
line = reviewers_file.readline()
while line != '':
line = line.split(',')
file = line[0]
values = [float(v) for v in line[24:31]]
values = ["%.2f" % round(v, 2) for v in values]
if file in file_values_map:
file_values_map[file].append(values)
line = reviewers_file.readline()
output_file.write('file')