return prediction
def predict_batch_from_model(patches, model):
"""Predict which pixels are tumor.
input: patch: `batch_size`x256x256x3, rgb image
input: model: keras model
output: prediction: 256x256x1, per-pixel tumor probability
"""
predictions = model.predict(patches)
predictions = predictions[:, :, :, 1]
return predictions
train_generator = generate_tiles_fast(train_samples, 32, shuffle=False)
validation_generator = generate_tiles_fast(validation_samples,
32,
shuffle=False)
train_steps = len(train_samples) // BATCH_SIZE
validation_steps = len(validation_samples) // BATCH_SIZE
#train_predictions = model.predict_generator(train_generator, steps= len(train_samples)//BATCH_SIZE)
#validation_predictions = model.predict_generator(validation_generator,
# steps= len(validation_samples)//BATCH_SIZE,
# callbacks=[TQDMCallback()])
#print(validation_predictions.shape)
#train_predictions = np.argmax(predictions, axis=-1) #multiple categories
def predict_batch_from_model(patches, model):
"""Predict which pixels are tumor.
input: patch: `batch_size`x256x256x3, rgb image
input: model: keras model
output: prediction: 256x256x1, per-pixel tumor probability
"""
predictions = model.predict(patches)
predictions = predictions[:, :, :, 1]
return predictions
train_generator = generate_tiles_fast(train_samples.sample(32,
random_state=42),
32,
shuffle=True)
validation_generator = generate_tiles_fast(validation_samples.sample(
32, random_state=42),
32,
shuffle=True)
filepath = "fast-allsamples-keras-improvement-{epoch:02d}-{val_acc:.2f}.hdf"
checkpoint = ModelCheckpoint(filepath,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
callbacks_list = [checkpoint]
model.fit_generator(train_generator,