def main(args):
#设定图像路径、答案路径、问题路径
image_dir = args.input_dir + '/Resized_Images/%s/'
annotation_file = args.input_dir + '/Annotations/v2_mscoco_%s_annotations.json'
question_file = args.input_dir + '/Questions/v2_OpenEnded_mscoco_%s_questions.json'
vocab_answer_file = args.output_dir + '/vocab_answers.txt'
answer_dict = text_processing.VocabDict(vocab_answer_file) #建立类
valid_answer_set = set(answer_dict.word_list) #载入文件得到单词列表并将其作为有效的回答集合
#分别对4类文件夹做处理
train = vqa_processing(image_dir, annotation_file, question_file,
valid_answer_set, 'train2014')
valid = vqa_processing(image_dir, annotation_file, question_file,
valid_answer_set, 'val2014')
test = vqa_processing(image_dir, annotation_file, question_file,
valid_answer_set, 'test2015')
test_dev = vqa_processing(image_dir, annotation_file, question_file,
valid_answer_set, 'test-dev2015')
#保存整合后的数据集到npy文件
np.save(args.output_dir + '/train.npy', np.array(train))
np.save(args.output_dir + '/valid.npy', np.array(valid))
np.save(args.output_dir + '/train_valid.npy', np.array(train + valid))
np.save(args.output_dir + '/test.npy', np.array(test))
np.save(args.output_dir + '/test-dev.npy', np.array(test_dev))
def __init__(self,
input_dir,
input_vqa,
max_qst_length=30,
max_num_ans=10,
transform=None):
self.input_dir = input_dir
self.vqa = np.load(input_dir + '/' + input_vqa)
self.qst_vocab = text_helper.VocabDict(input_dir +
'/vocab_questions.txt')
self.ans_vocab = text_helper.VocabDict(input_dir +
'/vocab_answers.txt')
self.max_qst_length = max_qst_length
self.max_num_ans = max_num_ans
self.load_ans = ('valid_answers'
in self.vqa[0]) and (self.vqa[0]['valid_answers']
is not None)
self.transform = transform
def main(args):
image_dir = args.input_dir+'/Resized_Images/%s/'
annotation_file = args.input_dir+'/Annotations/v2_mscoco_%s_annotations.json'
question_file = args.input_dir+'/Questions/v2_OpenEnded_mscoco_%s_questions.json'
vocab_answer_file = args.output_dir+'/vocab_answers.txt'
answer_dict = text_processing.VocabDict(vocab_answer_file)
valid_answer_set = set(answer_dict.word_list)
train = vqa_processing(image_dir, annotation_file, question_file, valid_answer_set, 'train2014')
valid = vqa_processing(image_dir, annotation_file, question_file, valid_answer_set, 'val2014')
test = vqa_processing(image_dir, annotation_file, question_file, valid_answer_set, 'test2015')
test_dev = vqa_processing(image_dir, annotation_file, question_file, valid_answer_set, 'test-dev2015')
np.save(args.output_dir+'/train.npy', np.array(train))
np.save(args.output_dir+'/valid.npy', np.array(valid))
np.save(args.output_dir+'/train_valid.npy', np.array(train+valid))
np.save(args.output_dir+'/test.npy', np.array(test))
np.save(args.output_dir+'/test-dev.npy', np.array(test_dev))