(x_train, y_train), (x_valid, y_valid), _ = load_training_data(task_idx=3,
output_size=224,
num_partitions=num_folds,
idx_partition=k_fold)
debug_visualize = False
if debug_visualize:
x_train = x_train[:32]
y_train = y_train[:32]
x_valid = x_valid[:32]
y_valid = y_valid[:32]
bv = BatchVisualization(images=x_train, true_labels=y_train)
bv()
num_classes = y_train.shape[1]
callbacks = config_cls_callbacks(run_name)
model = backbone(backbone_name, **backbone_options).classification_model(
input_shape=x_train.shape[1:],
num_classes=num_classes,
num_dense_layers=num_dense_layers,
num_dense_units=num_dense_units,
pooling=pooling,
dropout_rate=dropout_rate,
kernel_regularizer=dense_layer_regularizer,
save_to=run_name,
_, (x, y_true), _ = load_training_data(task_idx=task_idx, output_size=224)
if len(y_true.shape) == 3:
y_true = y_true[..., None]
if y_true[0].max() > 1:
y_true = (y_true > 127.5).astype(np.uint8)
model = backbone(backbone_name).segmentation_model(load_from=run_name, lr=0.001)
# max_num_images = 32
max_num_images = x.shape[0]
x = x[:max_num_images]
y_true = y_true[:max_num_images]
y_pred = model.predict(x, batch_size=8)
if task_idx == 1:
y_pred = task1_post_process(y_prediction=y_pred, threshold=0.5, gauss_sigma=2.)
mean_jaccard, thresholded_jaccard = compute_jaccard(y_true=y_true, y_pred=y_pred)
print('Mean jaccard = %.3f, Thresholded Jaccard = %.3f ' % (mean_jaccard, thresholded_jaccard))
bv = BatchVisualization(images=x,
true_masks=y_true,
pred_masks=y_pred)
bv()
# scores = model.evaluate(x, y_true, batch_size=32, verbose=1)
# print(scores)
y_valid = y_valid.astype(np.uint8)
n_samples_train = x_train.shape[0]
n_samples_valid = x_valid.shape[0]
debug_visualize = False
if debug_visualize:
x_train = x_train[:32]
y_train = y_train[:32]
x_valid = x_valid[:32]
y_valid = y_valid[:32]
bv = BatchVisualization(images=x_train, true_masks=y_train)
bv()
callbacks = config_seg_callbacks(run_name)
if from_run_name:
model = backbone(backbone_name).segmentation_model(
load_from=from_run_name, lr=init_lr)
else:
model = backbone(backbone_name,
**backbone_options).segmentation_model(
input_shape=x_train.shape[1:],
num_classes=y_train.shape[3],
upsampling_type=upsampling_type,
bottleneck=bottleneck,
init_nb_filters=init_nb_filters,
if __name__ == '__main__':
from datasets.ISIC2018 import *
from models import backbone
from misc_utils.visualization_utils import BatchVisualization
from misc_utils.eval_utils import get_confusion_matrix, get_precision_recall
backbone_name = 'inception_v3'
k_fold = 0
version = '0'
run_name = 'task3_' + backbone_name + '_k' + str(k_fold) + '_v' + version
_, (x, y_true), _ = load_training_data(task_idx=3,
output_size=224,
idx_partition=k_fold)
model = backbone(backbone_name).classification_model(load_from=run_name)
# max_num_images = 32
max_num_images = x.shape[0]
x = x[:max_num_images]
y_true = y_true[:max_num_images]
y_pred = model.predict(x)
_ = get_confusion_matrix(y_true=y_true, y_pred=y_pred, print_cm=True)
get_precision_recall(y_true=y_true, y_pred=y_pred)
bv = BatchVisualization(images=x, true_labels=y_true, pred_labels=y_pred)
bv()
