def run(config):
# Load options
file_path = config.file_path
g_path = config.g_path
d_path = config.d_path
# Initial values
g_reapeat_num = 6
d_reapeat_num = 6
c_dim = 2
g_conv_dim = 64
image_size = 256
d_conv_dim = 64
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Make generator netwrok and disciminator
G = Generator(g_conv_dim, c_dim, g_reapeat_num).to(device)
D = Discriminator(image_size, d_conv_dim, c_dim, d_reapeat_num).to(device)
# Load Networks
G.load(g_path)
D.load(d_path)
# Load X-ray image
image = Image.open(file_path)
w, h = image.size
image = np.array(image.getdata()).reshape(w, h, 3)
image = image.mean(axis=2).reshape(1, 1, w, h)
image = normalizatrion(image)
input_image = torch.from_numpy(image).type(torch.FloatTensor).to(device)
# Generate target vector
Target_normal = torch.tensor(np.array([[[0], [1]]]))
Target_normal = Target_normal.type(torch.FloatTensor).to(device)
# Synthesis to normal image
Normal_image = G(input_image, Target_normal)
# Make lesionmap
# calculate difference input and output
lesion_map = input_image - Normal_image[0]
lesion_map = to_image(lesion_map[0], c=1)
lesion_map = cv2.applyColorMap(lesion_map, cv2.COLORMAP_JET)
input_img = to_image(input_image[0], c=3)
normal_img = to_image(Normal_image[0], c=3)
synthesis_img = (input_image * 0.7 + lesion_map * 0.3).astype(np.uint8)
subject_image = cv2.hconcat(
[input_img, lesion_map, normal_img, synthesis_img])
plt.imshow(subject_image)
plt.show()
