def get_questions_processed_path(self, dataset_name=None):
if dataset_name is None:
dataset_name = self.get_current_dataset()
path = get_path(self.evaluation_type, dataset_name, 'questions_processed')
return path
def main(starting_counter):
test_args, test_exps = [], []
path = get_path('validation', 'abstract_scenes_v1', 'questions')
test_args.append([29994, 6, path])
test_exps.append(6)
test_args.append([29900, 32, path])
test_exps.append(32)
test_args.append([20000, 20, path])
test_exps.append(20)
tests_basis.create_tests([test_fn] * len(test_args), test_args, test_exps)
return tests_basis.main_tester("Testing questions batch loading",
starting_counter)
def main(starting_counter):
test_args, test_exps = [], []
path = get_path('validation', 'abstract_scenes_v1', 'annotations')
question_ids = [275780, 275781, 275782, 255060, 255061, 255062]
test_args.append([question_ids, path])
test_exps.append(len(question_ids))
test_args.append([question_ids[:1], path])
test_exps.append(1)
test_args.append([question_ids[:5], path])
test_exps.append(5)
tests_basis.create_tests([test_fn] * len(test_args), test_args, test_exps)
return tests_basis.main_tester("Testing annotations batch loading",
starting_counter)
def main(starting_counter):
test_args, test_exps = [], []
path = get_path('validation', 'abstract_scenes_v1', 'images')
image_ids = [
20000, 20001, 20002, 20003, 20004, 20005, 20006, 20007, 20008, 20009
]
test_args.append([image_ids, path])
test_exps.append(len(image_ids))
test_args.append([image_ids[:1], path])
test_exps.append(1)
test_args.append([image_ids[:5], path])
test_exps.append(5)
tests_basis.create_tests([test_fn] * len(test_args), test_args, test_exps)
return tests_basis.main_tester("Testing image batch loading",
starting_counter)
def _get_all_directory_paths():
images_directory_paths = []
features_directory_paths = []
images_directory_paths.append(
data_path.get_path('training', 'balanced_real_images', 'images'))
images_directory_paths.append(
data_path.get_path('training', 'balanced_binary_abstract_scenes',
'images'))
images_directory_paths.append(
data_path.get_path('training', 'abstract_scenes_v1', 'images'))
images_directory_paths.append(
data_path.get_path('validation', 'balanced_real_images', 'images'))
images_directory_paths.append(
data_path.get_path('validation', 'balanced_binary_abstract_scenes',
'images'))
images_directory_paths.append(
data_path.get_path('validation', 'abstract_scenes_v1', 'images'))
features_directory_paths.append(
data_path.get_path('training', 'balanced_real_images',
'images_features'))
features_directory_paths.append(
data_path.get_path('training', 'balanced_binary_abstract_scenes',
'images_features'))
features_directory_paths.append(
data_path.get_path('training', 'abstract_scenes_v1',
'images_features'))
features_directory_paths.append(
data_path.get_path('validation', 'balanced_real_images',
'images_features'))
features_directory_paths.append(
data_path.get_path('validation', 'balanced_binary_abstract_scenes',
'images_features'))
features_directory_paths.append(
data_path.get_path('validation', 'abstract_scenes_v1',
'images_features'))
return images_directory_paths, features_directory_paths
def get_img_features_path(self, dataset_name=None):
if dataset_name is None:
dataset_name = self.get_current_dataset()
return get_path(self.evaluation_type, dataset_name, 'images_features')
def get_annotations_path(self, dataset_name=None):
if dataset_name is None:
dataset_name = self.get_current_dataset()
return get_path(self.evaluation_type, dataset_name, 'annotations')
def get_img_path(self):
return get_path(self.evaluation_type, self.get_current_dataset(), 'images')
def get_top_answers_map():
global TOP_ANSWERS_MAP, TOP_ANSWERS_LIST, CLASS_WEIGHT
if os.path.exists(TOP_ANSWERS_PATH):
with open(TOP_ANSWERS_PATH, 'rb') as fp:
TOP_ANSWERS_MAP, TOP_ANSWERS_LIST, CLASS_WEIGHT = pickle.load(fp)
return TOP_ANSWERS_MAP, TOP_ANSWERS_LIST, CLASS_WEIGHT
top_answers_dict = {}
# get_annotations()[1] will return a dictionary of question id and multiple choice answer for this question
annotations_abstract_v1 = _get_annotations(
get_path('training', 'abstract_scenes_v1', 'annotations'))[1]
annotations_balanced_binary_abstract = _get_annotations(
get_path('training', 'balanced_binary_abstract_scenes',
'annotations'))[1]
annotations_balanced_real = _get_annotations(
get_path('training', 'balanced_real_images', 'annotations'))[1]
all_annotations = [
annotations_abstract_v1, annotations_balanced_binary_abstract,
annotations_balanced_real
]
for annot_dict in all_annotations:
for key, multiple_choice_answer in annot_dict.items():
if multiple_choice_answer in top_answers_dict:
top_answers_dict[multiple_choice_answer] += 1
else:
top_answers_dict[multiple_choice_answer] = 1
# return 2 columns array sorted on the second column, the first column is the answer and the second column is the count
sorted_top_answers = sorted(top_answers_dict.items(),
key=operator.itemgetter(1),
reverse=True)
# return the first column of the sorted_top_answers, that are the words
if TOP_ANSWERS_COUNT > len(sorted_top_answers):
raise ValueError(
"Top answers count is more than the number of answers !\n TOP_ANSWERS_COUNT = "
+ TOP_ANSWERS_COUNT)
TOP_ANSWERS_LIST = ([
row[0] for row in sorted_top_answers[:TOP_ANSWERS_COUNT]
])
TOP_ANSWERS_MAP = {}
for i in range(len(TOP_ANSWERS_LIST)):
ans = TOP_ANSWERS_LIST[i]
TOP_ANSWERS_MAP[ans] = i
CLASS_WEIGHT = _calculate_class_weights(TOP_ANSWERS_LIST, top_answers_dict)
# Write map and list to file
with open(TOP_ANSWERS_PATH, 'wb') as fp:
pickle.dump([TOP_ANSWERS_MAP, TOP_ANSWERS_LIST, CLASS_WEIGHT], fp)
return TOP_ANSWERS_MAP, TOP_ANSWERS_LIST, CLASS_WEIGHT
