def __init__(self):
self.dataSetFile = ''
self.successLogger = Logger(
os.path.join(FACE_DETECTION_ROOT, 'faceDetection.success'))
self.errorLogger = Logger(
os.path.join(FACE_DETECTION_ROOT, 'faceDetection.error'))
self.boundingboxFile = Logger(
os.path.join(FACE_DETECTION_ROOT, 'boundingbox.list'))
self.cc = cv2.CascadeClassifier(
os.path.join(FACE_DETECTION_ROOT,
'haarcascade_frontalface_alt.xml'))
def detectLandmarks(boundingboxList):
"""
detect landmarks in `src` and store the result in `dst`
"""
#bboxes = []
#landmarks = []
fl = Landmarker()
logger = Logger(os.path.join(FACE_ALIGNMENT_ROOT, 'landmark.list'))
# create bbox list
fid = open(boundingboxList, 'r');
fLines = fid.read().splitlines()
fid.close()
for line in fLines:
word = line.split()
filename = word[0]
img = cv2.imread(filename)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
bbox = BBox([int(word[1]), int(word[2]), int(word[3]), int(word[4])])\
.subBBox(0.1, 0.9, 0.2, 1)
landmark, status = fl.detectLandmark(gray, bbox)
'''
get real landmark position
'''
landmark = bbox.reprojectLandmark(landmark)
logger.writeMsg("%s" % filename)
for x, y in landmark:
logger.writeMsg(" %s %s" % (str(x), str(y)))
logger.writeMsg('\n')
'''
free memory: force the Garbage Collector to release
'''
gc.collect()
"transition":
namedtuple('transition',
('state', 'action', 'reward', 'next_state', 'done'))
})
print(f"Environment: {params['env_name']}\n"
f"Number of actions: {params['n_actions']}")
if params["do_intro_env"]:
intro_env()
env = make_atari(params["env_name"], params["seed"])
agent = Agent(**params)
experiment = Experiment()
logger = Logger(agent, experiment=experiment, **params)
if not params["train_from_scratch"]:
chekpoint = logger.load_weights()
agent.online_model.load_state_dict(
chekpoint["online_model_state_dict"])
agent.hard_update_target_network()
params.update({"beta": chekpoint["beta"]})
min_episode = chekpoint["episode"]
print("Keep training from previous run.")
else:
min_episode = 0
print("Train from scratch.")
if params["do_train"]:
