imgs_list.append(np.array(Image.open(image).resize((384, 384))))
masks_list.append(np.array(Image.open(mask).resize((384, 384))))
imgs_np = np.asarray(imgs_list)
masks_np = np.asarray(masks_list)
# %%
print(imgs_np.shape, masks_np.shape)
# %% [markdown]
# ## Plot images + masks + overlay (mask over original)
# %%
from keras_unet.utils import plot_imgs
plot_imgs(org_imgs=imgs_np, mask_imgs=masks_np, nm_img_to_plot=10, figsize=6)
# %% [markdown]
# ## Get data into correct shape, dtype and range (0.0-1.0)
# %%
print(imgs_np.max(), masks_np.max())
# %%
x = np.asarray(imgs_np, dtype=np.float32) / 255
y = np.asarray(masks_np, dtype=np.float32) / 255
# %%
print(x.max(), y.max())
# %%
model.save_weights(model_filename_final)
# Plot training history
if val_ratio != 0:
plot_segm_history(history)
else:
plot_segm_history(history, metrics=["iou"], losses=["loss"])
####################
# PLOT TEST RESULT #
####################
# Plot original + pred + overlay (pred on top of original)
images_test, file_names = load_test_data(test_directory)
# Predictions from best model according to val_loss
if val_ratio != 0:
model.load_weights(model_filename_best)
labels_pred_best = model.predict(images_test)
plot_imgs(org_imgs=images_test,
mask_imgs=labels_pred_best,
nm_img_to_plot=10)
# Predictions from final model
model.load_weights(model_filename_final)
labels_pred_final = model.predict(images_test)
plot_imgs(org_imgs=images_test, mask_imgs=labels_pred_final, nm_img_to_plot=10)
# Save final predictions
save_test_predictions(labels_pred_final, predictions_directory, file_names)
data_gen_args=dict(rotation_range=15.,
width_shift_range=0.05,
height_shift_range=0.05,
shear_range=50,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True,
fill_mode='constant'))
# Train model
results = model.fit_generator(
train_gen,
validation_data=([img_val_data, annot_val_data]),
steps_per_epoch=100,
epochs=epochs,
callbacks=[early_stopper, checkpointer, reduce_learning_rate])
if evaluate:
# Load testing data
x_test = np.load(directory +
r'\test_imgs_v256_3npy.npy').astype('float32')
y_test = np.load(directory + r'\test_annot_v5.npy').astype('float32')
# Getting the predictions for the testing set
y_pred = model.predict(x_test)
plot_imgs(org_imgs=x_test[500:, :, :, 1],
mask_imgs=y_test[500:, :, :, 1],
pred_imgs=y_pred[500:, :, :, 1],
nm_img_to_plot=5)
#%% Test predict
from keras_unet.utils import plot_imgs
model.load_weights(model_filename)
x_val, y_val = next(train_gen_patch)
print("train shape:", x_val.shape)
print("mask shape:", y_val.shape)
pred = model.predict(x_val)
print("pred shape:", pred.shape)
x_val = x_val[:,:,:,:3]
plot_imgs(
org_imgs=x_val, # required - original images
mask_imgs=y_val, # required - ground truth masks
pred_imgs=pred, # optional - predicted masks
nm_img_to_plot=x_val.shape[0], # optional - number of images to plot
color="red",
fontsize=25
)
#%%
from imgaug import augmenters as iaa
import skimage.io as io
import matplotlib.pyplot as plt
seq = iaa.Sequential([
iaa.Crop(px=(0, 16)), # crop images from each side by 0 to 16px (randomly chosen)
iaa.Fliplr(0.5), # horizontally flip 50% of the images
iaa.GaussianBlur(sigma=(0, 3.0)) # blur images with a sigma of 0 to 3.0