def generator_val():
while True:
for it in range(0, len(val_data_names), batch_size):
val_data_X, val_data_Y = load_images(
val_data_names[it:it + batch_size],
division=8,
crop=b_crop,
rescale=b_rescale,
scale=b_scale,
b_debug=b_debug,
normcv2=b_normcv2,
rows=rows,
fulldepth=fulldepth,
cols=cols,
removeBackground=removeBackground,
equalize=equalize)
yield val_data_X, val_data_Y
def generator():
random.shuffle(train_data_names)
while True:
for it in range(0, len(train_data_names), batch_size):
X, Y = load_images(train_data_names[it:it + batch_size],
division=8,
b_debug=b_debug,
crop=b_crop,
rescale=b_rescale,
scale=b_scale,
normcv2=b_normcv2,
fulldepth=fulldepth,
rows=rows,
cols=cols,
removeBackground=removeBackground,
equalize=equalize)
yield X, Y
seq = test_data_names[image]['image'].split('\\')[-3]
frame = 0
if checkBadFrame(seq, frame):
print 'skipped', seq, frame
continue
if test_data_names[image]['face'] == 0:
print 'SKIP NO GT', seq, frame
continue
t = time.time()
test_data_X, _ = load_images(test_data_names[image:image + 1],
crop=b_crop,
rescale=b_rescale,
scale=b_scale,
b_debug=False,
normcv2=b_normcv2,
rows=rows,
fulldepth=False,
cols=cols,
equalize=True,
removeBackground=True,
division=4)
pred = model.predict(x=test_data_X, batch_size=batch_size, verbose=0)
for index, i in enumerate(pred):
gt_head = bool(test_data_names[image + index]['face'])
if gt_head:
gt_coord = test_data_names[image + index]['facecord']
head = False
img = cv2.imread(test_data_names[image + index]['image'],
cv2.IMREAD_ANYDEPTH)
img = cv2.resize(img, (cols, rows))