def test_softencoding():
image_path = "Dataset/Test/images/0.JPEG"
image = cv2.imread(image_path)
image_small = cv2.resize(image, (16, 16))
l, a, b = convert_rgb_to_lab(image)
l_small, a_small, b_small = convert_rgb_to_lab(image_small)
encoding = softencoding(a_small, b_small)
nonzeros_per_pixels = np.count_nonzero(encoding, axis=2)
num_diff_five = np.where(nonzeros_per_pixels != 5)
assert (num_diff_five[0].size == 0)
def __getitem__(self, idx):
i = idx * self.batch_size
length = min(self.batch_size, (len(self.names) - i))
batch_x = np.empty((length, 64, 64, 1), dtype=np.float32)
batch_y = np.empty((length, 16, 16, 313), dtype=np.float32)
for i_batch in range(length):
name = self.names[i]
image = os.path.join(self.image_folder, name)
bgr = cv.imread(image)
l, a, b = convert_rgb_to_lab(bgr)
x = l / 255.
out_a = cv.resize(a, (16, 16), cv.INTER_CUBIC)
out_b = cv.resize(b, (16, 16), cv.INTER_CUBIC)
out_a = out_a.astype(np.int32) - 128 ## / 256 * 190 - 90
out_b = out_b.astype(np.int32) - 128 ## / 256 * 190 - 90
y = softencoding(out_a, out_b)
batch_x[i_batch, :, :, 0] = x
batch_y[i_batch] = y
i += 1
return batch_x, batch_y
def pixel_analyzing(src):
cord = np.load("pts_in_hull.npy")
nbrs = NearestNeighbors(n_neighbors=1, algorithm='ball_tree').fit(cord)
main = os.fsencode(src)
result = np.zeros(313)
for folder in os.listdir(main):
foldername = src + "/" + os.fsdecode(folder) + "/images/"
for file in os.listdir(os.fsencode(foldername)):
filename = os.fsdecode(file)
if filename.endswith(
('.JPEG', '.png', '.jpg')
) and not filename.startswith("."): # image extension we need
image = cv2.imread(foldername + filename)
l_channel, a_channel, b_channel = convert_rgb_to_lab(image)
a = a_channel.reshape(-1, 1)
b = b_channel.reshape(-1, 1)
X = np.column_stack((a, b))
distances, indices = nbrs.kneighbors(X)
result[indices] += 1 #saved the indices in an array
result = result / np.sum(result)
sigma = 5
wts = np.exp(-result**2 / (2 * sigma**2))
wts = wts / np.sum([wts], axis=1)[:, np.newaxis]
#print(wts)
#print(np.sum(wts))
np.save("prior_probs", arr=wts)
def test_decode():
### TODO: find a better way?
image_path = "Dataset/Test/images/0.JPEG"
image = cv2.imread(image_path)
image_small = cv2.resize(image, (16, 16))
l, a, b = convert_rgb_to_lab(image)
l_small, a_small, b_small = convert_rgb_to_lab(image_small)
a_small = a_small.astype(np.int32) - 128
b_small = b_small.astype(np.int32) - 128
encoding = softencoding(a_small, b_small)
a_dec, b_dec = decode(encoding)
new_img_path = "test_res0.JPEG"
merge_channels((new_img_path, l.astype(dtype=np.uint8)),
(new_img_path, a_dec.astype(dtype=np.uint8),
b_dec.astype(dtype=np.uint8)), "Dataset/Test",
"unit_testing", True)
import cv2
## Change with wanted model name
model_name = 'new_model.h5'
net = Network(model_name)
net.load()
test_folder = "Dataset/Test/images"
folder = os.fsencode(test_folder)
for file in os.listdir(folder):
filename = os.fsdecode(file)
if filename.endswith(('.JPEG', '.png', '.jpg')):
image = cv2.imread(os.path.join(test_folder, filename))
image_small = cv2.resize(image, (16, 16))
l, a, b = convert_rgb_to_lab(image)
x_test = np.empty((1, 64, 64, 1), dtype=np.float32)
x_test[0, :, :, 0] = l / 255.
pred = net.predict(x_test)
a, b = decode(pred.reshape((16, 16, 313)))
new_result_name = model_name + "_" + filename
merge_channels((new_result_name, l.astype(dtype=np.uint8)),
(new_result_name, a.astype(dtype=np.uint8),
b.astype(dtype=np.uint8)), "Dataset", "merged", True)