image_shape_x_coarse = (64,64,128)
image_shape_coarse = (64,64,3)
mask_shape_coarse = (64,64,1)
label_shape_coarse = (64,64,1)
img_shape_g = (64,64,3)
ndf=64
ncf=128
nff=128
## Load models
K.clear_session()
opt = Adam()
g_local_model = fine_generator(x_coarse_shape=image_shape_x_coarse,input_shape=image_shape_fine,mask_shape=mask_shape_fine,nff=nff)
g_local_model.load_weights(args.weight_name_local)
g_local_model.compile(loss='mse', optimizer=opt)
g_global_model = coarse_generator(img_shape=image_shape_coarse,mask_shape=mask_shape_coarse,ncf=ncf)
g_global_model.load_weights(args.weight_name_global)
g_global_model.compile(loss='mse',optimizer=opt)
## Create Output Directory
out_path = args.out_dir
directories = [out_path,out_path+'/Coarse',out_path+'/Fine']
for directory in directories:
if not os.path.exists(directory):
os.makedirs(directory)
## Find file numbers,paths or names
if args.test_data == 'DRIVE':
limit = 20
elif args.test_data == 'CHASE':
d_model3 = discriminator(image_shape_coarse,
label_shape_coarse,
ndf,
n_layers,
n_downsampling=0,
name="D3") # D1 Coarse
d_model4 = discriminator(image_shape_coarse,
label_shape_coarse,
ndf,
n_layers,
n_downsampling=1,
name="D4") # D2 Coarse
# define generator models
g_coarse_model = coarse_generator(img_shape=image_shape_coarse,
n_downsampling=2,
n_blocks=9,
n_channels=1)
g_fine_model = fine_generator(x_coarse_shape=image_shape_xglobal,
input_shape=image_shape_fine,
nff=nff,
n_blocks=3)
# define fundus2angio
gan_model = aagan(g_fine_model, g_coarse_model, d_model1, d_model2,
d_model3, d_model4, image_shape_fine, image_shape_coarse,
image_shape_xglobal, label_shape_fine,
label_shape_coarse)
# train model
train(d_model1,
d_model2,
label_shape_fine,
ndf,
name="D1")
d_model2 = discriminator_ae(image_shape_coarse,
label_shape_coarse,
ndf,
name="D2")
g_model_fine = fine_generator(x_coarse_shape=image_shape_xglobal,
input_shape=image_shape_fine,
mask_shape=mask_shape_fine,
nff=nff,
n_blocks=3)
g_model_coarse = coarse_generator(img_shape=image_shape_coarse,
mask_shape=mask_shape_coarse,
n_downsampling=2,
n_blocks=9,
ncf=ncf,
n_channels=1)
if args.resume_training == 'yes':
#weight_name_global = "global_model_000070.h5"
g_model_coarse.load_weights(args.weight_name_global)
#weight_name_local = "local_model_000070.h5"
g_model_fine.load_weights(args.weight_name_local)
rvgan_model = RVgan(g_model_fine, g_model_coarse, d_model1, d_model2,
image_shape_fine, image_shape_coarse,
image_shape_xglobal, mask_shape_fine,
mask_shape_coarse, label_shape_fine,
label_shape_coarse, args.inner_weight)