def im_test(im):
face_info = lib.align(im[:, :, (2, 1, 0)], front_face_detector,
lmark_predictor)
# Samples
if len(face_info) == 0:
logging.warning('No faces are detected.')
prob = -1 # we ignore this case
else:
# Check how many faces in an image
logging.info('{} faces are detected.'.format(len(face_info)))
max_prob = -1
# If one face is fake, the image is fake
for _, point in face_info:
rois = []
for i in range(sample_num):
roi, _ = lib.cut_head([im], point, i)
rois.append(cv2.resize(roi[0], tuple(cfg.IMG_SIZE[:2])))
vis_im(rois, 'tmp/vis.jpg')
prob = solver.test(rois)
prob = np.mean(
np.sort(prob[:, 0])[np.round(sample_num / 2).astype(int):])
if prob >= max_prob:
max_prob = prob
prob = max_prob
return prob
def main(args):
# Parse config
cfg_file = args.cfg
with open(cfg_file, 'r') as f:
cfg = edict(yaml.load(f))
# Data generator
data_gen = DataProcTest(
face_img_dir=args.data_dir,
cache_path=args.cache_path,
sample_num=args.sample_num)
with tf.Session() as sess:
# Build network
reso_net = ResoNet(cfg=cfg, is_train=False)
reso_net.build()
# Build solver
solver = Solver(sess=sess, cfg=cfg, net=reso_net)
solver.init()
for i in range(data_gen.batch_num):
data = data_gen.get_batch(i, resize=cfg.IMG_SIZE[:2])
images = data['images']
name_list = data['name_list']
vis_im(images, 'tmp/vis.jpg')
prob = solver.test(images)
print(np.mean(prob[:, 0]))
def main(args):
# Parse config
cfg_file = args.cfg
with open(cfg_file, 'r') as f:
cfg = edict(yaml.load(f))
# Data generator
data_gen = DataProcTrain(
face_img_dir=args.data_dir,
cache_path=args.cache_path,
anno_path=args.list,
batch_size=cfg.TRAIN.BATCH_SIZE,
is_shuffle=True)
cfg.TRAIN.DECAY_STEP = data_gen.batch_num
epoch = cfg.TRAIN.NUM_EPOCH
with tf.Session() as sess:
# Build network
reso_net = ResoNet(cfg=cfg, is_train=True)
reso_net.build()
# Init solver
solver = Solver(sess=sess, cfg=cfg, net=reso_net)
solver.init()
# Train
count = 0
for epoch_id in range(epoch):
for i in range(data_gen.batch_num):
data = data_gen.get_batch(i, resize=cfg.IMG_SIZE[:2])
images = data['images']
labels = data['images_label']
ims_tmp = vis_im(images, 'tmp/vis.jpg')
summary, prob, net_loss, total_loss, weights = solver.train(images, labels)
pred_labels = np.argmax(prob, axis=1)
print('====================================')
print('Net loss: {}'.format(net_loss))
print('Total loss: {}'.format(total_loss))
print('Real label: {}'.format(np.array(labels)))
print('Pred label: {}'.format(pred_labels))
print('Neg hard mining: {}'.format(weights))
print('epoch: {}, batch_idx: {}'.format(epoch_id, i))
if count % 100 == 0:
solver.writer.add_summary(summary, count)
count += 1
if epoch_id % 2 == 0:
solver.save(epoch_id)