def process():
def _parse_image_id(image):
return int(image.split('.')[0].split('_')[-1])
model = AttentionModel()
ans2top_ans = AnswerTokenToTopAnswer()
task_data_dir = '/usr/data/fl302/code/utils/bs_data_maker'
task_data_file = os.path.join(task_data_dir, 'task_data_for_verif.json')
task_data = load_json(task_data_file)
is_valid = []
num = len(task_data)
for i, info in enumerate(task_data):
print('%d/%d' % (i, num))
image = info['image']
image_id = _parse_image_id(image)
question = info['target']
answer = info['answer']
scores = model.inference(image_id, question)
scores[:, -1] = -10.
# pdb.set_trace()
top_ans_id = ans2top_ans.direct_query(answer)
if top_ans_id == 2000:
raise Exception('Warning: answer oov')
scores = scores.flatten()
pred_top_ans_id = scores.argmax()
is_valid.append(int(pred_top_ans_id == top_ans_id))
n_valid = sum(is_valid)
print('valid: %d/%d' % (n_valid, num))
save_json(os.path.join(task_data_dir, 'task_data_verif_state.json'),
is_valid)
def __init__(self, ckpt_file='model/v1_vqa_VQA/v1_vqa_VQA_best2/model.ckpt-135000',
use_dis_reward=False):
self.g = tf.Graph()
self.ckpt_file = ckpt_file
from models.vqa_soft_attention import AttentionModel
from vqa_config import ModelConfig
config = ModelConfig()
self.ans2id = AnswerTokenToTopAnswer()
self.use_dis_reward = use_dis_reward
with self.g.as_default():
self.sess = tf.Session()
self.model = AttentionModel(config, phase='test_broadcast')
self.model.build()
vars = tf.trainable_variables()
self.saver = tf.train.Saver(var_list=vars)
self.saver.restore(self.sess, ckpt_file)
def __init__(self, ckpt_file='/usr/data/fl302/code/inverse_vqa/model/mlb_attention_v2/model.ckpt-170000',
use_dis_reward=False):
self.g = tf.Graph()
self.ckpt_file = ckpt_file
self.v1tov2 = TopAnswerVersionConverter()
from models.vqa_soft_attention_v2 import AttentionModel
from vqa_config import ModelConfig
config = ModelConfig()
self.ans2id = AnswerTokenToTopAnswer()
self.use_dis_reward = use_dis_reward
with self.g.as_default():
self.sess = tf.Session()
self.model = AttentionModel(config, phase='test_broadcast')
self.model.build()
vars = tf.trainable_variables()
self.saver = tf.train.Saver(var_list=vars)
self.saver.restore(self.sess, ckpt_file)
def __init__(self, ckpt_file='', use_dis_reward=False,
use_attention_model=False):
self.g = tf.Graph()
self.ckpt_file = ckpt_file
self.use_attention_model = use_attention_model
from models.vqa_base import BaseModel
from vqa_config import ModelConfig
config = ModelConfig()
self.ans2id = AnswerTokenToTopAnswer()
self.use_dis_reward = use_dis_reward
with self.g.as_default():
self.sess = tf.Session()
if self.use_attention_model:
self.model = AttentionModel(config, phase='test')
self.model.build()
else:
self.model = BaseModel(config, phase='test')
self.model.build()
vars = tf.trainable_variables()
self.saver = tf.train.Saver(var_list=vars)
self.saver.restore(self.sess, ckpt_file)
def __init__(self, use_dis_reward=False):
self.g = tf.Graph()
self.ans2id = AnswerTokenToTopAnswer()
self.use_dis_reward = use_dis_reward
self.model = None