def investigate_dataset_size(db, feature_extraction_methods, args, proportions,
best_params, config_path, size_invest_path):
config_dict = {}
for feat_extr_meth, par, best_param in zip(feature_extraction_methods,
args, best_params):
arg_dict = {list(par.keys())[0]: best_param}
if par["const_args"] is not None:
arg_dict.update(par["const_args"])
accuracy_values = []
for proportion in proportions:
X_train, y_train, X_test, y_test = train_and_test_split(
db, proportion)
recognizer = FaceRecognizer(feat_extr_meth, arg_dict)
recognizer.fit(X_train, y_train)
accuracy = find_accuracy(X_test, y_test, recognizer)
accuracy_values.append(accuracy)
method_name = feat_extr_meth.__name__
method_dict = {
method_name: {
'best_param': proportions[np.argmax(accuracy_values)],
'best_accuracy': max(accuracy_values)
}
}
config_dict.update(method_dict)
with open(config_path, 'w') as f:
yaml.dump(config_dict, f)
train_imgs_count = list(map(lambda x: int(x * 10), proportions))
file_path = os.path.join(size_invest_path, method_name)
make_plot(train_imgs_count, accuracy_values, file_path, method_name,
"Dataset size investigation")
print(f"Dataset size investigation for {method_name} is done")
def person_recognition_example(db, proportion, feature_extraction_methods,
args, images_path):
X_train, y_train, X_test, y_test = train_and_test_split(db, proportion)
[img_width, img_height] = X_test[0].shape
new_width, new_height = img_width * 3, img_height * 3
resized_img = cv2.resize(X_test[0], (new_width, new_height))
show_image(resized_img, "Original image")
cv2.imwrite(os.path.join(images_path, "Original image.png"), X_test[0])
for feat_extr_meth, par in zip(feature_extraction_methods, args):
recognizer = FaceRecognizer(feat_extr_meth, par)
recognizer.fit(X_train, y_train)
prediction, pred_number = recognizer.predict(X_test[0])
method_name = feat_extr_meth.__name__
resized_img = cv2.resize(X_train[pred_number], (new_width, new_height))
show_image(resized_img, method_name)
cv2.imwrite(os.path.join(images_path, f"{method_name}.png"),
X_train[pred_number])
def investigate_params(X_train, y_train, X_test, y_test,
feature_extraction_methods, args, param_graphs_path,
config_path):
best_params = []
config_dict = {}
for feat_extr_meth, par in zip(feature_extraction_methods, args):
accuracy_values = []
for arg in list(par.values())[0]:
arg_dict = {list(par.keys())[0]: arg}
if par["const_args"] is not None:
arg_dict.update(par["const_args"])
recognizer = FaceRecognizer(feat_extr_meth, arg_dict)
recognizer.fit(X_train, y_train)
accuracy = find_accuracy(X_test, y_test, recognizer)
accuracy_values.append(accuracy)
best_param = list(par.values())[0][np.argmax(accuracy_values)]
best_params.append(best_param)
method_name = feat_extr_meth.__name__
method_dict = {
method_name: {
'best_param': best_param,
'best_accuracy': max(accuracy_values)
}
}
config_dict.update(method_dict)
with open(config_path, 'w') as f:
yaml.dump(config_dict, f)
paameters = list(par.values())[0]
file_path = os.path.join(param_graphs_path, method_name)
make_plot(paameters, accuracy_values, file_path, method_name,
"Parameters investigation")
print(f"Parameters investigation for {method_name} is done")
return best_params
