def predict(self, test_dir: str, dataset_name: str = "", save: bool = True):
"""Evaluates a new set of images using the trained CNN.
Args:
test_dir: Relative path to the validation directory (e.g., 'dataset/test').
dataset_name: Dataset descriptive name.
save: Save results to an Excel file.
"""
# Configure loading and pre-processing functions
print('Reading test data...')
test_datagen = ImageDataGenerator(preprocessing_function=self._preprocessing_function)
test_generator = test_datagen.flow_from_directory(
test_dir,
target_size=self._target_size,
batch_size=1, # A batch size of 1 ensures that all test images are processed
class_mode='categorical',
shuffle=False
)
# Predict categories
predictions = self._model.predict_generator(test_generator, steps=test_generator.samples)
predicted_labels = np.argmax(predictions, axis=1).ravel().tolist()
# Format results and compute classification statistics
results = Results(test_generator.class_indices, dataset_name=dataset_name)
accuracy, confusion_matrix, classification = results.compute(test_generator.filenames, test_generator.classes,
predicted_labels)
# Display and save results
results.print(accuracy, confusion_matrix)
if save:
results.save(confusion_matrix, classification)
def predict(self, dataset: List[str], dataset_name: str = "", save: bool = True) -> \
Tuple[float, np.ndarray, List[Tuple[str, str, str]]]:
"""Evaluates a new set of images using the trained classifier.
Args:
dataset: Paths to the test images.
dataset_name: Dataset descriptive name.
save: Save results to an Excel file.
Returns:
Classification accuracy.
Confusion matrix.
Detailed per image classification results.
"""
# Extract features
test_desc = []
test_labels = []
for path in dataset:
test_desc.extend(self._extract_bow_features(path))
test_labels.append(self._labels.get(Dataset.get_label(path)))
# Predict categories
predicted_labels = (self._classifier.predict(np.array(test_desc, np.float32))[1]).ravel().tolist()
predicted_labels = [int(label) for label in predicted_labels]
# Format results and compute classification statistics
results = Results(self._labels, dataset_name=dataset_name)
accuracy, confusion_matrix, classification = results.compute(dataset, test_labels, predicted_labels)
results.print(accuracy, confusion_matrix)
if save:
results.save(confusion_matrix, classification)
return accuracy, confusion_matrix, classification
