def keras_model_fn(hyperparameters):
'''
hyperparameters: The hyperparameters passed to SageMaker TrainingJob that runs your TensorFlow training
script. You can use this to pass hyperparameters to your training script.
'''
# Logic to do the following:
# 1. Instantiate the Keras model
# 2. Compile the keras model
config = hyperparameters
model = get_model_and_load_weights(config)
compile_model(model, config)
return model
def manip(args, test_list, u_model):
if args.test_weights_path == '':
weights_path = os.path.join(args.check_dir, args.output_name + '_model_' + args.time + '.hdf5')
else:
weights_path = os.path.join(args.data_root_dir, args.test_weights_path)
output_dir = os.path.join(args.data_root_dir, 'results', args.net)
manip_out_dir = os.path.join(output_dir, 'manip_output')
try:
safe_mkdir(manip_out_dir)
except:
pass
# Compile the loaded model
manip_model = compile_model(args=args, uncomp_model=u_model)
try:
manip_model.load_weights(weights_path)
except:
raise NotImplementedError('Unable to find weights path.')
# Manipulating capsule vectors
print('Testing... This will take some time...')
for i, img in enumerate(tqdm(test_list)):
sitk_img = sitk.ReadImage(os.path.join(args.data_root_dir, 'imgs', img[0]))
img_data = sitk.GetArrayFromImage(sitk_img)
num_slices = img_data.shape[0]
sitk_mask = sitk.ReadImage(os.path.join(args.data_root_dir, 'masks', img[0]))
gt_data = sitk.GetArrayFromImage(sitk_mask)
x, y = img_data[num_slices//2, :, :], gt_data[num_slices//2, :, :]
x, y = np.expand_dims(np.expand_dims(x, -1), 0), np.expand_dims(np.expand_dims(y, -1), 0)
noise = np.zeros([1, 512, 512, 1, 16])
x_recons = []
for dim in trange(16):
for r in [-0.25, -0.125, 0, 0.125, 0.25]:
tmp = np.copy(noise)
tmp[:, :, :, :, dim] = r
x_recon = manip_model.predict([x, y, tmp])
x_recons.append(x_recon)
x_recons = np.concatenate(x_recons)
out_img = combine_images(x_recons, height=16)
out_image = out_img * 4096
out_image[out_image > 574] = 574
out_image = out_image / 574 * 255
Image.fromarray(out_image.astype(np.uint8)).save(os.path.join(manip_out_dir, img[0][:-4] + '_manip_output.png'))
print('Done.')
def test(args, u_model, val_samples, val_shape, test_samples, test_shape):
out_dir = os.path.join(args.root_dir, 'predictions', args.exp_name,
args.net)
try:
safe_mkdir(out_dir)
except:
pass
# Compile the loaded model
model = compile_model(args=args, uncomp_model=u_model)
# Load testing weights
if args.test_weights_path != '':
try:
model.load_weights(args.test_weights_path)
output_filename = os.path.join(
out_dir,
os.path.basename(args.test_weights_path)[:-5] + '.csv')
except Exception as e:
print(e)
raise NotImplementedError('Failed to load weights file in test.py')
else:
try:
model.load_weights(
os.path.join(
args.check_dir,
args.output_name + '_model_' + args.time + '.hdf5'))
output_filename = os.path.join(
out_dir, args.output_name + '_model_' + args.time + '.csv')
except Exception as e:
print(e)
raise NotImplementedError('Failed to load weights from training.')
test_datagen = ImageDataGenerator(samplewise_center=True,
samplewise_std_normalization=True,
rescale=1. / 255)
if args.thresh_level == 0.:
thresh = 0.5
else:
thresh = args.thresh_level
# TESTING SECTION
y_true_test = []
def test_data_gen(gen):
while True:
batch = gen.next()
y_true_test.append(batch[1][0])
x_batch = np.copy(batch[0])
for i, x in enumerate(batch[0]):
x2 = np.copy(x)
x2 = x2 + abs(np.min(x2))
x2 /= (np.max(x2) + 1e-7)
x2 = (x2 - 0.5) * 2.
x_batch[i, ...] = x2
yield x_batch
test_flow_gen = test_datagen.flow_from_directory(os.path.join(
args.img_dir, 'test'),
target_size=test_shape,
class_mode='binary',
batch_size=1,
seed=12,
shuffle=False)
filenames = np.asarray(test_flow_gen.filenames)
test_flow_gen.reset()
test_gen = test_data_gen(test_flow_gen)
results = model.predict_generator(test_gen,
max_queue_size=1,
workers=1,
use_multiprocessing=False,
steps=test_samples,
verbose=args.verbose)
if args.net.find('caps') != -1:
test_scores = results[0]
reconstructions = results[1]
else:
test_scores = results
polyp_ids = []
for f in tqdm(filenames, desc='Loading filenames'):
temp = os.path.basename(f).split('_')
try:
polyp_ids.append('m_{}_{}'.format(
os.path.dirname(f)[2:], temp[1][:6]))
except:
polyp_ids.append('m_{}_{}'.format(
os.path.dirname(f)[2:], temp[0][:6]))
unique_polyp_results_ALL = []
unique_polyp_results_NBI = []
unique_polyp_results_NBIF = []
unique_polyp_results_NBIN = []
unique_polyp_results_WL = []
unique_polyp_results_WLF = []
unique_polyp_results_WLN = []
unique_polyp_results_NEAR = []
unique_polyp_results_FAR = []
unique_polyp_labels = []
unique_polyp_names = []
counts = Counter(polyp_ids)
for s, num in tqdm(counts.items(), desc='Computing Scores'):
current_polyp_results_ALL = []
current_polyp_results_NBI = []
current_polyp_results_NBIF = []
current_polyp_results_NBIN = []
current_polyp_results_WL = []
current_polyp_results_WLF = []
current_polyp_results_WLN = []
current_polyp_results_NEAR = []
current_polyp_results_FAR = []
current_polyp_name = s
for _ in range(1, num + 1): # loop over all images of same polyp
pos = polyp_ids.index(s)
current_image_score = test_scores[pos][0]
current_polyp_results_ALL.append(current_image_score)
current_filename = os.path.basename(filenames[pos])
split_name = current_filename.split('-')
if len(split_name) < 4:
print('Encountered improperly named image. Please fix: {}.'.
format(current_filename))
continue
if split_name[3] == 'NBI':
current_polyp_results_NBI.append(current_image_score)
elif split_name[3] == 'NBIF':
current_polyp_results_NBIF.append(current_image_score)
current_polyp_results_NBI.append(current_image_score)
current_polyp_results_FAR.append(current_image_score)
elif split_name[3] == 'NBIN':
current_polyp_results_NBIN.append(current_image_score)
current_polyp_results_NBI.append(current_image_score)
current_polyp_results_NEAR.append(current_image_score)
elif split_name[3] == 'WL':
current_polyp_results_WL.append(current_image_score)
elif split_name[3] == 'WLF':
current_polyp_results_WLF.append(current_image_score)
current_polyp_results_WL.append(current_image_score)
current_polyp_results_FAR.append(current_image_score)
elif split_name[3] == 'WLN':
current_polyp_results_WLN.append(current_image_score)
current_polyp_results_WL.append(current_image_score)
current_polyp_results_NEAR.append(current_image_score)
else:
Warning('Encountered unexpected imaging type: {}.'.format(
split_name[3]))
polyp_ids[pos] = s + '_c' # mark the image as seen
unique_polyp_names.append(current_polyp_name)
unique_polyp_results_ALL.append(
np.mean(np.asarray(current_polyp_results_ALL)))
if current_polyp_results_NBI:
unique_polyp_results_NBI.append(
np.mean(np.asarray(current_polyp_results_NBI)))
else:
unique_polyp_results_NBI.append(np.nan)
if current_polyp_results_NBIF:
unique_polyp_results_NBIF.append(
np.mean(np.asarray(current_polyp_results_NBIF)))
else:
unique_polyp_results_NBIF.append(np.nan)
if current_polyp_results_NBIN:
unique_polyp_results_NBIN.append(
np.mean(np.asarray(current_polyp_results_NBIN)))
else:
unique_polyp_results_NBIN.append(np.nan)
if current_polyp_results_WL:
unique_polyp_results_WL.append(
np.mean(np.asarray(current_polyp_results_WL)))
else:
unique_polyp_results_WL.append(np.nan)
if current_polyp_results_WLF:
unique_polyp_results_WLF.append(
np.mean(np.asarray(current_polyp_results_WLF)))
else:
unique_polyp_results_WLF.append(np.nan)
if current_polyp_results_WLN:
unique_polyp_results_WLN.append(
np.mean(np.asarray(current_polyp_results_WLN)))
else:
unique_polyp_results_WLN.append(np.nan)
if current_polyp_results_NEAR:
unique_polyp_results_NEAR.append(
np.mean(np.asarray(current_polyp_results_NEAR)))
else:
unique_polyp_results_NEAR.append(np.nan)
if current_polyp_results_FAR:
unique_polyp_results_FAR.append(
np.mean(np.asarray(current_polyp_results_FAR)))
else:
unique_polyp_results_FAR.append(np.nan)
unique_polyp_labels = np.asarray(unique_polyp_labels)
warnings.filterwarnings("ignore")
predictions_IMAGES = np.where(test_scores > thresh, 1., 0.)
predictions_ALL = np.where(unique_polyp_results_ALL > thresh, 1., 0.)
try:
predictions_ALL[np.argwhere(
np.isnan(unique_polyp_results_ALL))] = np.nan
except:
predictions_ALL = np.asarray(unique_polyp_results_ALL)
predictions_NBI = np.where(unique_polyp_results_NBI > thresh, 1., 0.)
try:
predictions_NBI[np.argwhere(
np.isnan(unique_polyp_results_NBI))] = np.nan
except:
predictions_NBI = np.asarray(unique_polyp_results_NBI)
predictions_NBIF = np.where(unique_polyp_results_NBIF > thresh, 1., 0.)
try:
predictions_NBIF[np.argwhere(
np.isnan(unique_polyp_results_NBIF))] = np.nan
except:
predictions_NBIF = np.asarray(unique_polyp_results_NBIF)
predictions_NBIN = np.where(unique_polyp_results_NBIN > thresh, 1., 0.)
try:
predictions_NBIN[np.argwhere(
np.isnan(unique_polyp_results_NBIN))] = np.nan
except:
predictions_NBIN = np.asarray(unique_polyp_results_NBIN)
predictions_WL = np.where(unique_polyp_results_WL > thresh, 1., 0.)
try:
predictions_WL[np.argwhere(np.isnan(unique_polyp_results_WL))] = np.nan
except:
predictions_WL = np.asarray(unique_polyp_results_WL)
predictions_WLF = np.where(unique_polyp_results_WLF > thresh, 1., 0.)
try:
predictions_WLF[np.argwhere(
np.isnan(unique_polyp_results_WLF))] = np.nan
except:
predictions_WLF = np.asarray(unique_polyp_results_WLF)
predictions_WLN = np.where(unique_polyp_results_WLN > thresh, 1., 0.)
try:
predictions_WLN[np.argwhere(
np.isnan(unique_polyp_results_WLN))] = np.nan
except:
predictions_WLN = np.asarray(unique_polyp_results_WLN)
predictions_NEAR = np.where(unique_polyp_results_NEAR > thresh, 1., 0.)
try:
predictions_NEAR[np.argwhere(
np.isnan(unique_polyp_results_NEAR))] = np.nan
except:
predictions_NEAR = np.asarray(unique_polyp_results_NEAR)
predictions_FAR = np.where(unique_polyp_results_FAR > thresh, 1., 0.)
try:
predictions_FAR[np.argwhere(
np.isnan(unique_polyp_results_FAR))] = np.nan
except:
predictions_FAR = np.asarray(unique_polyp_results_FAR)
warnings.resetwarnings()
np.savetxt(output_filename,
np.stack([
predictions_IMAGES,
np.squeeze(predictions_ALL[np.argwhere(
np.isfinite(unique_polyp_results_ALL))]),
np.squeeze(predictions_NBI[np.argwhere(
np.isfinite(unique_polyp_results_NBI))]),
np.squeeze(predictions_NBIF[np.argwhere(
np.isfinite(unique_polyp_results_NBIF))]),
np.squeeze(predictions_NBIN[np.argwhere(
np.isfinite(unique_polyp_results_NBIN))]),
np.squeeze(predictions_WL[np.argwhere(
np.isfinite(unique_polyp_results_WL))]),
np.squeeze(predictions_WLF[np.argwhere(
np.isfinite(unique_polyp_results_WLF))]),
np.squeeze(predictions_WLN[np.argwhere(
np.isfinite(unique_polyp_results_WLN))]),
np.squeeze(predictions_NEAR[np.argwhere(
np.isfinite(unique_polyp_results_NEAR))]),
np.squeeze(predictions_FAR[np.argwhere(
np.isfinite(unique_polyp_results_FAR))])
],
axis=0),
delimiter=',')
def manip(args, u_model, test_samples):
out_dir = join(args.data_root_dir, 'results', args.exp_name, args.net)
try:
makedirs(out_dir)
except:
pass
out_img_dir = join(out_dir, 'manip_output')
try:
makedirs(out_img_dir)
except:
pass
# Compile the loaded model
model = compile_model(args=args, uncomp_model=u_model)
# Load testing weights
if args.test_weights_path != '':
try:
model.load_weights(args.test_weights_path)
out_name = basename(args.test_weights_path)[:-5]
except Exception as e:
print(e)
raise Exception('Failed to load weights from training.')
else:
try:
model.load_weights(
join(args.check_dir,
args.output_name + '_model_' + args.time + '.hdf5'))
out_name = args.output_name + '_model_' + args.time
except Exception as e:
print(e)
raise Exception('Failed to load weights from training.')
x_test = normalize_img(
np.expand_dims(test_samples[0], axis=-1).astype(np.float32))
if args.num_classes == 1:
y_test = np.expand_dims(test_samples[2][:, 25],
axis=-1) # 25 should be avg mal score
y_test[y_test < 3.] = 0.
y_test[y_test >= 3.] = 1.
else:
y_test = to_categorical(np.rint(test_samples[2][:, -2]) - 1)
print('Creating manipulated outputs.')
for mal_val in trange(y_test.shape[1]):
index = np.argmax(y_test, 1) == mal_val
number = np.random.randint(low=0, high=sum(index) - 1)
x, y = x_test[index][number], y_test[index][number]
x, y = np.expand_dims(x, 0), np.expand_dims(y, 0)
if args.net.find('xcaps') != -1:
noise = np.zeros([1, 6, 16])
elif args.net == 'capsnet':
noise = np.zeros([1, y_test.shape[1], 16])
else:
raise NotImplementedError(
'Specified Network does not have proper implementation in manip.py'
)
x_recons = []
for attr in range(noise.shape[0]):
for dim in range(16):
for r in [
-0.5, -0.25, -0.2, -0.15, -0.1, -0.05, 0, 0.05, 0.1,
0.15, 0.2, 0.25, 0.5
]:
tmp = np.copy(noise)
tmp[attr, :, dim] = r
if args.net.find('xcaps') != -1:
x_recon = model.predict([x, tmp])
elif args.net == 'capsnet':
x_recon = model.predict([x, y, tmp])
else:
raise NotImplementedError(
'Specified Network does not have proper implementation in manip.py'
)
x_recons.append(x_recon[-1])
x_recons = np.concatenate(x_recons)
img = combine_images(x_recons, height=16)
pil_img = Image.fromarray(255 * img).convert('L')
pil_img.save(
join(out_img_dir,
out_name + '_{}_{}.png'.format(attr, mal_val + 1)))
def test(args, u_model, test_samples):
out_dir = os.path.join(args.data_root_dir, 'results', args.exp_name,
args.net)
safe_mkdir(out_dir)
out_img_dir = os.path.join(out_dir, 'recons')
safe_mkdir(out_img_dir)
# Compile the loaded model
model = compile_model(args=args, uncomp_model=u_model)
# Load testing weights
if args.test_weights_path != '':
output_filename = os.path.join(
out_dir, 'results_' +
os.path.basename(args.test_weights_path)[:-5] + '.csv')
try:
model.load_weights(args.test_weights_path)
except Exception as e:
print(e)
raise Exception('Failed to load weights from training.')
else:
output_filename = os.path.join(
out_dir,
'results_' + args.output_name + '_model_' + args.time + '.csv')
try:
model.load_weights(
os.path.join(
args.check_dir,
args.output_name + '_model_' + args.time + '.hdf5'))
except Exception as e:
print(e)
raise Exception('Failed to load weights from training.')
test_datagen = ImageDataGenerator(samplewise_center=False,
samplewise_std_normalization=False,
rescale=None)
# TESTING SECTION
def data_gen(gen):
while True:
x, y = gen.next()
yield x, y
x_test = normalize_img(
np.expand_dims(test_samples[0], axis=-1).astype(np.float32))
if args.num_classes == 1:
y_test = np.rint(test_samples[2][:, -2])
else:
y_test = get_pseudo_label([1., 2., 3., 4., 5.], test_samples[2][:, -2],
test_samples[2][:, -1])
test_gen = data_gen(
test_datagen.flow(x=x_test,
y=y_test,
batch_size=1,
shuffle=False,
seed=12))
results = model.predict_generator(test_gen,
max_queue_size=1,
workers=1,
use_multiprocessing=False,
steps=len(x_test),
verbose=1)
if args.net.find('caps') != -1:
y_pred = results[0]
x_recon = results[-1]
img = combine_images(np.concatenate([x_test[:250:5], x_recon[:250:5]]))
pil_img = Image.fromarray(255 * img).convert('L')
if args.test_weights_path != '':
img_filename = os.path.join(
out_img_dir,
os.path.basename(args.test_weights_path)[:-5] +
'_real_and_recon.png')
else:
img_filename = os.path.join(
out_img_dir, args.output_name + '_model_' + args.time +
'_real_and_recon.png')
pil_img.save(os.path.join(out_img_dir, img_filename))
else:
y_pred = results
if args.num_classes == 1:
gt = y_test
pred = np.squeeze(np.rint(y_pred * 4 + 1))
else:
gt = np.argmax(y_test, axis=1) + 1
pred = np.argmax(y_pred, axis=1) + 1
if args.num_classes == 1:
cmat = confusion_matrix(gt, pred, labels=[1, 2, 3, 4, 5])
test_acc_cat, test_acc_all = compute_within_one_acc(cmat)
test_acc_cat_weighted, test_acc_all_weighted = np.zeros_like(
test_acc_cat), np.zeros_like(test_acc_all)
else:
cmat = confusion_matrix(gt, pred, labels=[1, 2, 3, 4, 5])
test_acc_cat, test_acc_all = compute_within_one_acc(cmat)
cmat_weighted = confusion_matrix(
gt,
pred,
labels=[1, 2, 3, 4, 5],
sample_weight=1. / np.var([1., 2., 3., 4., 5.] * y_pred, axis=1))
test_acc_cat_weighted, test_acc_all_weighted = compute_within_one_acc(
cmat_weighted)
with open(output_filename, 'w', newline='') as f:
fw = csv.writer(f, delimiter=',')
fw.writerow(
['Malignancy Accuracy', 'Malignancy Accuracy Confidence Weighted'])
fw.writerow([
'{:05f}'.format(test_acc_all),
'{:05f}'.format(test_acc_all_weighted)
])
fw.writerow(['Malignancy Accuracy by Score:'])
fw.writerow(['{:05f}'.format(num) for num in test_acc_cat])
fw.writerow(['Malignancy Accuracy by Score Confidence Weighted:'])
fw.writerow(['{:05f}'.format(num) for num in test_acc_cat_weighted])
fw.writerow(['Malignancy Confusion Matrix:'])
for row in cmat:
fw.writerow(['{:05f}'.format(num) for num in row])
if args.net.find('dcaps') != -1 or args.net == 'xcapsnet':
if args.net.find('simple') != -1 or args.net == 'xcapsnet':
attr_pred = np.rint(
np.swapaxes(np.asarray(results[1:-1]), 0, -1) * 4 + 1)
y_attr = np.rint(
np.concatenate(
(np.expand_dims(test_samples[2][:, -18], axis=-1),
np.expand_dims(test_samples[2][:, -12], axis=-1),
np.expand_dims(test_samples[2][:, -10], axis=-1),
np.expand_dims(test_samples[2][:, -8], axis=-1),
np.expand_dims(test_samples[2][:, -6], axis=-1),
np.expand_dims(test_samples[2][:, -4], axis=-1)),
axis=1)).astype(np.int64)
for i in range(y_attr.shape[1]):
attr_cmat = confusion_matrix(y_attr[:, i],
attr_pred[:, i],
labels=[1, 2, 3, 4, 5])
class_acc, total_acc = compute_within_one_acc(attr_cmat)
fw.writerow(['Attribute {} Accuracy'.format(i)])
fw.writerow(['{:05f}'.format(total_acc)])
fw.writerow(['Attribute {} Accuracy by Score:'.format(i)])
fw.writerow(['{:05f}'.format(num) for num in class_acc])
fw.writerow(['Attribute {} Confusion Matrix:'.format(i)])
for row in attr_cmat:
fw.writerow(['{:05f}'.format(num) for num in row])
else:
attr_pred = np.argmax(
np.rollaxis(np.asarray(results[1:-1]), 0, -1), axis=-1) + 1
y_attr = np.concatenate(
(np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -18],
test_samples[2][:, -17]),
axis=1),
np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -12],
test_samples[2][:, -11]),
axis=1),
np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -10],
test_samples[2][:, -9]),
axis=1),
np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -8],
test_samples[2][:, -7]),
axis=1),
np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -6],
test_samples[2][:, -5]),
axis=1),
np.expand_dims(get_pseudo_label([1., 2., 3., 4., 5.],
test_samples[2][:, -4],
test_samples[2][:, -3]),
axis=1)),
axis=1)
gt_attr = np.argmax(y_attr, axis=2) + 1
for i in range(gt_attr.shape[1]):
attr_cmat = confusion_matrix(gt_attr[:, i],
attr_pred[:, i],
labels=[1, 2, 3, 4, 5])
attr_cmat_weighted = confusion_matrix(
gt_attr[:, i],
attr_pred[:, i],
labels=[1, 2, 3, 4, 5],
sample_weight=1. /
np.var([1., 2., 3., 4., 5.] * np.rollaxis(
np.asarray(results[1:-1]), 0, -1)[:, i],
axis=1))
class_acc, total_acc = compute_within_one_acc(attr_cmat)
class_acc_weighted, total_acc_weighted = compute_within_one_acc(
attr_cmat_weighted)
fw.writerow([
'Attribute {} Accuracy'.format(i),
'Attribute {} Accuracy Confidence Weighted'.format(i)
])
fw.writerow([
'{:05f}'.format(total_acc),
'{:05f}'.format(total_acc_weighted)
])
fw.writerow(['Attribute {} Accuracy by Score:'.format(i)])
fw.writerow(['{:05f}'.format(num) for num in class_acc])
fw.writerow([
'Attribute {} Accuracy by Score Confidence Weighted:'.
format(i)
])
fw.writerow(
['{:05f}'.format(num) for num in class_acc_weighted])
fw.writerow(['Attribute {} Confusion Matrix:'.format(i)])
for row in attr_cmat:
fw.writerow(['{:05f}'.format(num) for num in row])
def test(args, u_model, val_samples, val_shape, test_samples, test_shape):
out_dir = os.path.join(args.root_dir, 'results', args.exp_name, args.net)
try:
safe_mkdir(out_dir)
except:
pass
# Compile the loaded model
model = compile_model(args=args, uncomp_model=u_model)
# Load testing weights
if args.test_weights_path != '':
try:
model.load_weights(args.test_weights_path)
output_filename = os.path.join(
out_dir,
os.path.basename(args.test_weights_path)[:-5] + '.csv')
except Exception as e:
print(e)
raise NotImplementedError('Failed to load weights file in test.py')
else:
try:
model.load_weights(
os.path.join(
args.check_dir,
args.output_name + '_model_' + args.time + '.hdf5'))
output_filename = os.path.join(
out_dir, args.output_name + '_model_' + args.time + '.csv')
except Exception as e:
print(e)
raise NotImplementedError('Failed to load weights from training.')
test_datagen = ImageDataGenerator(samplewise_center=True,
samplewise_std_normalization=True,
rescale=1. / 255)
# VALIDATION SECTION
if args.thresh_level == 0.:
# We use this section to choose a threshold which maximizes the harmonic mean between sensitivity and specificity.
y_true_val = []
def val_data_gen(gen):
while True:
batch = gen.next()
y_true_val.append(batch[1][0])
x_batch = np.copy(batch[0])
for i, x in enumerate(batch[0]):
x2 = np.copy(x)
x2 = x2 + abs(np.min(x2))
x2 /= (np.max(x2) + 1e-7)
x2 = (x2 - 0.5) * 2.
x_batch[i, ...] = x2
yield x_batch
val_flow_gen = test_datagen.flow_from_directory(os.path.join(
args.img_dir, 'val'),
target_size=val_shape,
class_mode='binary',
batch_size=1,
seed=12,
shuffle=False)
val_flow_gen.reset()
val_gen = val_data_gen(val_flow_gen)
val_results = model.predict_generator(val_gen,
max_queue_size=1,
workers=1,
use_multiprocessing=False,
steps=val_samples,
verbose=args.verbose)
if args.net.find('caps') != -1:
val_scores = val_results[0]
val_reconstructions = val_results[1]
else:
val_scores = val_results
val_y_true = np.asarray(
y_true_val[:-(len(y_true_val) - len(val_flow_gen.filenames))])
thresh, [val_acc, val_sen,
val_spec] = find_thresh_level(val_scores, val_y_true,
'pseudof1')
else:
thresh = args.thresh_level
# TESTING SECTION
y_true_test = []
def test_data_gen(gen):
while True:
batch = gen.next()
y_true_test.append(batch[1][0])
x_batch = np.copy(batch[0])
for i, x in enumerate(batch[0]):
x2 = np.copy(x)
x2 = x2 + abs(np.min(x2))
x2 /= (np.max(x2) + 1e-7)
x2 = (x2 - 0.5) * 2.
x_batch[i, ...] = x2
yield x_batch
test_flow_gen = test_datagen.flow_from_directory(os.path.join(
args.img_dir, 'test'),
target_size=test_shape,
class_mode='binary',
batch_size=1,
seed=12,
shuffle=False)
filenames = np.asarray(test_flow_gen.filenames)
test_flow_gen.reset()
test_gen = test_data_gen(test_flow_gen)
results = model.predict_generator(test_gen,
max_queue_size=1,
workers=1,
use_multiprocessing=False,
steps=test_samples,
verbose=args.verbose)
if args.net.find('caps') != -1:
test_scores = results[0]
reconstructions = results[1]
else:
test_scores = results
test_y_true = np.asarray(
y_true_test[:-(len(y_true_test) - len(test_flow_gen.filenames))])
y_true_check = []
polyp_ids = []
for f in tqdm(filenames, desc='Loading filenames'):
y_true_check.append(f[0])
temp = os.path.basename(f).split('_')
try:
polyp_ids.append('m_{}_{}'.format(
os.path.dirname(f)[2:], temp[1][:6]))
except:
polyp_ids.append('m_{}_{}'.format(
os.path.dirname(f)[2:], temp[0][:6]))
y_true_check = np.asarray(y_true_check, dtype=np.float32)
assert np.array_equal(test_y_true, y_true_check), 'Error: Order of images and labels not preserved! ' \
'Cannot match images to labels.'
unique_polyp_results_ALL = []
unique_polyp_results_NBI = []
unique_polyp_results_NBIF = []
unique_polyp_results_NBIN = []
unique_polyp_results_WL = []
unique_polyp_results_WLF = []
unique_polyp_results_WLN = []
unique_polyp_results_NEAR = []
unique_polyp_results_FAR = []
unique_polyp_labels = []
unique_polyp_names = []
counts = Counter(polyp_ids)
for s, num in tqdm(counts.items(), desc='Computing Scores'):
current_polyp_results_ALL = []
current_polyp_results_NBI = []
current_polyp_results_NBIF = []
current_polyp_results_NBIN = []
current_polyp_results_WL = []
current_polyp_results_WLF = []
current_polyp_results_WLN = []
current_polyp_results_NEAR = []
current_polyp_results_FAR = []
current_polyp_name = s
current_polyp_label = test_y_true[polyp_ids.index(s)]
for _ in range(1, num + 1): # loop over all images of same polyp
pos = polyp_ids.index(s)
current_image_score = test_scores[pos][0]
current_polyp_results_ALL.append(current_image_score)
current_filename = os.path.basename(filenames[pos])
split_name = current_filename.split('-')
if len(split_name) < 4:
print('Encountered improperly named image. Please fix: {}.'.
format(current_filename))
continue
if split_name[3] == 'NBI':
current_polyp_results_NBI.append(current_image_score)
elif split_name[3] == 'NBIF':
current_polyp_results_NBIF.append(current_image_score)
current_polyp_results_NBI.append(current_image_score)
current_polyp_results_FAR.append(current_image_score)
elif split_name[3] == 'NBIN':
current_polyp_results_NBIN.append(current_image_score)
current_polyp_results_NBI.append(current_image_score)
current_polyp_results_NEAR.append(current_image_score)
elif split_name[3] == 'WL':
current_polyp_results_WL.append(current_image_score)
elif split_name[3] == 'WLF':
current_polyp_results_WLF.append(current_image_score)
current_polyp_results_WL.append(current_image_score)
current_polyp_results_FAR.append(current_image_score)
elif split_name[3] == 'WLN':
current_polyp_results_WLN.append(current_image_score)
current_polyp_results_WL.append(current_image_score)
current_polyp_results_NEAR.append(current_image_score)
else:
Warning('Encountered unexpected imaging type: {}.'.format(
split_name[3]))
polyp_ids[pos] = s + '_c' # mark the image as seen
unique_polyp_names.append(current_polyp_name)
unique_polyp_labels.append(current_polyp_label)
unique_polyp_results_ALL.append(
np.mean(np.asarray(current_polyp_results_ALL)))
if current_polyp_results_NBI:
unique_polyp_results_NBI.append(
np.mean(np.asarray(current_polyp_results_NBI)))
else:
unique_polyp_results_NBI.append(np.nan)
if current_polyp_results_NBIF:
unique_polyp_results_NBIF.append(
np.mean(np.asarray(current_polyp_results_NBIF)))
else:
unique_polyp_results_NBIF.append(np.nan)
if current_polyp_results_NBIN:
unique_polyp_results_NBIN.append(
np.mean(np.asarray(current_polyp_results_NBIN)))
else:
unique_polyp_results_NBIN.append(np.nan)
if current_polyp_results_WL:
unique_polyp_results_WL.append(
np.mean(np.asarray(current_polyp_results_WL)))
else:
unique_polyp_results_WL.append(np.nan)
if current_polyp_results_WLF:
unique_polyp_results_WLF.append(
np.mean(np.asarray(current_polyp_results_WLF)))
else:
unique_polyp_results_WLF.append(np.nan)
if current_polyp_results_WLN:
unique_polyp_results_WLN.append(
np.mean(np.asarray(current_polyp_results_WLN)))
else:
unique_polyp_results_WLN.append(np.nan)
if current_polyp_results_NEAR:
unique_polyp_results_NEAR.append(
np.mean(np.asarray(current_polyp_results_NEAR)))
else:
unique_polyp_results_NEAR.append(np.nan)
if current_polyp_results_FAR:
unique_polyp_results_FAR.append(
np.mean(np.asarray(current_polyp_results_FAR)))
else:
unique_polyp_results_FAR.append(np.nan)
unique_polyp_labels = np.asarray(unique_polyp_labels)
warnings.filterwarnings("ignore")
predictions_IMAGES = np.where(test_scores > thresh, 1., 0.)
predictions_ALL = np.where(unique_polyp_results_ALL > thresh, 1., 0.)
try:
predictions_ALL[np.argwhere(
np.isnan(unique_polyp_results_ALL))] = np.nan
except:
predictions_ALL = np.asarray(unique_polyp_results_ALL)
predictions_NBI = np.where(unique_polyp_results_NBI > thresh, 1., 0.)
try:
predictions_NBI[np.argwhere(
np.isnan(unique_polyp_results_NBI))] = np.nan
except:
predictions_NBI = np.asarray(unique_polyp_results_NBI)
predictions_NBIF = np.where(unique_polyp_results_NBIF > thresh, 1., 0.)
try:
predictions_NBIF[np.argwhere(
np.isnan(unique_polyp_results_NBIF))] = np.nan
except:
predictions_NBIF = np.asarray(unique_polyp_results_NBIF)
predictions_NBIN = np.where(unique_polyp_results_NBIN > thresh, 1., 0.)
try:
predictions_NBIN[np.argwhere(
np.isnan(unique_polyp_results_NBIN))] = np.nan
except:
predictions_NBIN = np.asarray(unique_polyp_results_NBIN)
predictions_WL = np.where(unique_polyp_results_WL > thresh, 1., 0.)
try:
predictions_WL[np.argwhere(np.isnan(unique_polyp_results_WL))] = np.nan
except:
predictions_WL = np.asarray(unique_polyp_results_WL)
predictions_WLF = np.where(unique_polyp_results_WLF > thresh, 1., 0.)
try:
predictions_WLF[np.argwhere(
np.isnan(unique_polyp_results_WLF))] = np.nan
except:
predictions_WLF = np.asarray(unique_polyp_results_WLF)
predictions_WLN = np.where(unique_polyp_results_WLN > thresh, 1., 0.)
try:
predictions_WLN[np.argwhere(
np.isnan(unique_polyp_results_WLN))] = np.nan
except:
predictions_WLN = np.asarray(unique_polyp_results_WLN)
predictions_NEAR = np.where(unique_polyp_results_NEAR > thresh, 1., 0.)
try:
predictions_NEAR[np.argwhere(
np.isnan(unique_polyp_results_NEAR))] = np.nan
except:
predictions_NEAR = np.asarray(unique_polyp_results_NEAR)
predictions_FAR = np.where(unique_polyp_results_FAR > thresh, 1., 0.)
try:
predictions_FAR[np.argwhere(
np.isnan(unique_polyp_results_FAR))] = np.nan
except:
predictions_FAR = np.asarray(unique_polyp_results_FAR)
warnings.resetwarnings()
scores_IMAGEWISE = compute_scores(y_true=test_y_true,
y_pred=predictions_IMAGES)
scores_ALL = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_ALL))]),
y_pred=np.squeeze(predictions_ALL[np.argwhere(
np.isfinite(unique_polyp_results_ALL))]))
scores_NBI = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_NBI))]),
y_pred=np.squeeze(predictions_NBI[np.argwhere(
np.isfinite(unique_polyp_results_NBI))]))
scores_NBIF = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_NBIF))]),
y_pred=np.squeeze(predictions_NBIF[np.argwhere(
np.isfinite(unique_polyp_results_NBIF))]))
scores_NBIN = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_NBIN))]),
y_pred=np.squeeze(predictions_NBIN[np.argwhere(
np.isfinite(unique_polyp_results_NBIN))]))
scores_WL = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_WL))]),
y_pred=np.squeeze(predictions_WL[np.argwhere(
np.isfinite(unique_polyp_results_WL))]))
scores_WLF = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_WLF))]),
y_pred=np.squeeze(predictions_WLF[np.argwhere(
np.isfinite(unique_polyp_results_WLF))]))
scores_WLN = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_WLN))]),
y_pred=np.squeeze(predictions_WLN[np.argwhere(
np.isfinite(unique_polyp_results_WLN))]))
scores_NEAR = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_NEAR))]),
y_pred=np.squeeze(predictions_NEAR[np.argwhere(
np.isfinite(unique_polyp_results_NEAR))]))
scores_FAR = compute_scores(
y_true=np.squeeze(unique_polyp_labels[np.argwhere(
np.isfinite(unique_polyp_results_FAR))]),
y_pred=np.squeeze(predictions_FAR[np.argwhere(
np.isfinite(unique_polyp_results_FAR))]))
np.savetxt(
output_filename,
np.stack([
scores_IMAGEWISE, scores_ALL, scores_NBI, scores_NBIF, scores_NBIN,
scores_WL, scores_WLF, scores_WLN, scores_NEAR, scores_FAR
],
axis=0),
delimiter=',')
print(
'- Testing Complete! Results on ALL Polyps -\nAccuracy: {}\nSensitivity: {}\nSpecificity: {}'
.format(scores_ALL[0], scores_ALL[1], scores_ALL[2]))
def train(args, u_model, train_samples, val_samples):
# Compile the loaded model
model = compile_model(args=args, uncomp_model=u_model)
# Load pre-trained weights
if args.finetune_weights_path != '':
try:
model.load_weights(args.finetune_weights_path)
except Exception as e:
print(e)
print(
'!!! Failed to load custom weights file. Training without pre-trained weights. !!!'
)
# Set the callbacks
callbacks = get_callbacks(args)
if args.aug_data:
train_datagen = ImageDataGenerator(
samplewise_center=False,
samplewise_std_normalization=False,
rotation_range=45,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.1,
zoom_range=0.1,
fill_mode='nearest',
horizontal_flip=True,
vertical_flip=True,
rescale=None,
preprocessing_function=custom_train_data_augmentation)
val_datagen = ImageDataGenerator(samplewise_center=False,
samplewise_std_normalization=False,
rescale=None)
else:
train_datagen = ImageDataGenerator(samplewise_center=False,
samplewise_std_normalization=False,
rotation_range=0,
width_shift_range=0.,
height_shift_range=0.,
shear_range=0.,
zoom_range=0.,
fill_mode='nearest',
horizontal_flip=False,
vertical_flip=False,
rescale=None)
val_datagen = ImageDataGenerator(samplewise_center=False,
samplewise_std_normalization=False,
rescale=None)
if debug:
save_dir = args.img_aug_dir
else:
save_dir = None
def xcaps_data_gen(gen):
while True:
x, y = gen.next()
if args.num_classes == 1:
mal = np.array([y[i][0][6, 0] for i in range(y.shape[0])])
else:
mal = np.array([y[i][0][6, 1:] for i in range(y.shape[0])])
yield x, [
mal,
np.array([y[i][0][0, 0] for i in range(y.shape[0])]),
np.array([y[i][0][1, 0] for i in range(y.shape[0])]),
np.array([y[i][0][2, 0] for i in range(y.shape[0])]),
np.array([y[i][0][3, 0] for i in range(y.shape[0])]),
np.array([y[i][0][4, 0] for i in range(y.shape[0])]),
np.array([y[i][0][5, 0] for i in range(y.shape[0])]),
x * np.expand_dims(
np.array([y[i][1] for i in range(y.shape[0])]), axis=-1)
]
def capsnet_data_gen(gen):
while True:
x, y = gen.next()
if args.num_classes == 1:
y = np.array([y[i][0][6, 0] for i in range(y.shape[0])])
else:
y = np.array([y[i][0][6, 1:] for i in range(y.shape[0])])
yield [x, y], [y, x]
# Prepare images and labels for training
train_imgs = normalize_img(
np.expand_dims(train_samples[0], axis=-1).astype(np.float32))
val_imgs = normalize_img(
np.expand_dims(val_samples[0], axis=-1).astype(np.float32))
train_labels = []
val_labels = []
n_attr = 9 # 8 attr + mal score
skip_attr_list = [1, 2]
for i in range(n_attr):
skip = False
if skip_attr_list:
for j in skip_attr_list:
if i == j: #indexing from negative side
skip_attr_list.remove(j)
skip = True
if args.num_classes == 1 and i == n_attr - 1:
tlab = np.repeat(np.expand_dims(train_samples[2][:,
2 * i + n_attr],
axis=-1),
6,
axis=1)
tlab[tlab < 3.] = 0.
tlab[tlab >= 3.] = 1.
train_labels.append(tlab)
vlab = np.repeat(np.expand_dims(val_samples[2][:, 2 * i + n_attr],
axis=-1),
6,
axis=1)
vlab[vlab < 3.] = 0.
vlab[vlab >= 3.] = 1.
val_labels.append(vlab)
skip = True
if not skip:
train_labels.append(
np.hstack(
(np.expand_dims(
(train_samples[2][:, 2 * i + n_attr] - 1) / 4.,
axis=-1),
get_pseudo_label([1., 2., 3., 4., 5.],
train_samples[2][:, 2 * i + n_attr],
train_samples[2][:,
2 * i + 1 + n_attr]))))
val_labels.append(
np.hstack(
(np.expand_dims(
(val_samples[2][:, 2 * i + n_attr] - 1) / 4., axis=-1),
get_pseudo_label([1., 2., 3., 4., 5.],
val_samples[2][:, 2 * i + n_attr],
val_samples[2][:, 2 * i + 1 + n_attr]))))
train_labels = np.rollaxis(np.asarray(train_labels), 0, 2)
val_labels = np.rollaxis(np.asarray(val_labels), 0, 2)
new_labels = np.empty((len(train_labels), 2), dtype=np.object)
for i in range(len(train_labels)):
new_labels[i, 0] = train_labels[i]
if args.masked_recon:
new_labels[i, 1] = train_samples[1][i]
else:
new_labels[i, 1] = np.ones_like(train_samples[1][i])
train_labels = new_labels
new_labels = np.empty((len(val_labels), 2), dtype=np.object)
for i in range(len(val_labels)):
new_labels[i, 0] = val_labels[i]
if args.masked_recon:
new_labels[i, 1] = val_samples[1][i]
else:
new_labels[i, 1] = np.ones_like(val_samples[1][i])
val_labels = new_labels
train_flow_gen = train_datagen.flow(x=train_imgs,
y=train_labels,
batch_size=args.batch_size,
shuffle=True,
seed=12,
save_to_dir=save_dir)
val_flow_gen = val_datagen.flow(x=val_imgs,
y=val_labels,
batch_size=args.batch_size,
shuffle=True,
seed=12,
save_to_dir=save_dir)
if args.net.find('xcaps') != -1:
train_gen = xcaps_data_gen(train_flow_gen)
val_gen = xcaps_data_gen(val_flow_gen)
elif args.net.find('capsnet') != -1:
train_gen = capsnet_data_gen(train_flow_gen)
val_gen = capsnet_data_gen(val_flow_gen)
else:
raise NotImplementedError(
'Data generator not found for specified network. Please check train.py file.'
)
# Settings
train_steps = len(train_samples[0]) // args.batch_size
val_steps = len(val_samples[0]) // args.batch_size
workers = 4
multiproc = True
# Run training
history = model.fit_generator(train_gen,
max_queue_size=40,
workers=workers,
use_multiprocessing=multiproc,
steps_per_epoch=train_steps,
validation_data=val_gen,
validation_steps=val_steps,
epochs=args.epochs,
class_weight=None,
callbacks=callbacks,
verbose=args.verbose,
shuffle=True)
# Plot the training data collected
plot_training(history, args)