def run_predict():
with tf.Graph().as_default(), tf.device('/cpu:0'):
parser = argparse.ArgumentParser()
parser.add_argument('--checkpoint_dir',
type=str,
default='./captcha_train',
help='Directory where to restore checkpoint.')
parser.add_argument('--captcha_dir',
type=str,
default='./captcha_dir',
help='Directory where to get captcha images.')
FLAGS, unparsed = parser.parse_known_args()
input_images, input_filenames = input_data(FLAGS.captcha_dir)
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
total_count = len(input_filenames)
true_count = 0.
if total_count == 1:
return str(text[0])
def run_predict(img_data):
with tf.Graph().as_default(), tf.device('/cpu:0'):
input_filenames = ''
# input_images, input_filenames = input_dir_data(FLAGS.captcha_dir)
input_images = input_img_data(img_data)
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
print(tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
total_count = len(input_filenames)
true_count = 0.
if total_count != 0:
for i in range(total_count):
print('image ' + input_filenames[i] + " recognize ----> '" +
text[i] + "'")
if text[i] in input_filenames[i]:
true_count += 1
precision = true_count / total_count
print('%s true/total: %d/%d recognize @ 1 = %.3f' %
(datetime.now(), true_count, total_count, precision))
elif total_count == 0:
print(text[0])
return text[0]
def predict(image_file):
with tf.Graph().as_default(), tf.device('/cpu:0'):
input_images = input_data(image_file)
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint('./captcha_train'))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
return text[0]
def run_predict(image_path):
with tf.Graph().as_default(), tf.device('/cpu:0'):
images, files = input_data(image_path)
images = tf.constant(images)
logits = captcha_model.inference(images, keep_prob=1)
result = captcha_model.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
print(tf.train.latest_checkpoint('./models'))
saver.restore(sess, tf.train.latest_checkpoint('./models'))
recog_result = sess.run(result)
sess.close()
print("recog_result: %s" % recog_result)
text = one_hot_to_text(recog_result[0])
print("recog text is: %s" % text)
return text
def run_predict(image_base64):
with tf.Graph().as_default(), tf.device('/cpu:0'):
input_images, input_filenames = input_data(image_base64)
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
# print(tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
total_count = len(input_filenames)
result = ""
for i in range(total_count):
result = text[i]
return result
def run_predict(url_yzm):
with tf.Graph().as_default(), tf.device('/cpu:0'):
input_images = [getImage(url_yzm)]
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint(checkpoint_dir))
print(tf.train.latest_checkpoint(checkpoint_dir))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
total_count = len(input_images)
true_count = 0.
for i in range(total_count):
print('image ' + input_images[i] + " recognize ----> '" + text[i] +
"'")
def run_predict():
with tf.Graph().as_default():
input_images, input_filenames = input_data(
FLAGS.captcha_dir) #得到文件夹内所有照片和文件名
epoches = len(input_images) // Batch_size
offset = len(input_images) - (epoches - 1) * Batch_size
images = tf.placeholder(tf.float32,
[Batch_size, IMAGE_HEIGHT * IMAGE_WIDTH],
name='input')
logits = captcha.inference(images, keep_prob=1, is_training=True)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session()
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
for each in range(epoches):
feed_dict = input_images[each * Batch_size:min(
(each + 1) * Batch_size, len(input_images))]
recog_result = sess.run(result, feed_dict={images: feed_dict})
text = one_hot_to_texts(recog_result)
total_count = len(feed_dict)
true_count = 0.
for i in range(total_count):
print('image ' + input_filenames[i + each * Batch_size] +
" recognize ----> '" + text[i] + "'")
with open('recognize.txt', 'a') as f:
f.write('image ' + input_filenames[i + each * Batch_size] +
'recognize ----> ' + text[i] + '\n')
if text[i] in input_filenames[i + each * Batch_size]:
true_count += 1
precision = true_count / total_count
print(
'%s epoch: %d ,true/total: %d/%d recognize @ = %.3f' %
(datetime.now(), each + 1, true_count, total_count, precision))
sess.close()
def run_predict():
with tf.Graph().as_default(), tf.device('/gpu:0'):
input_images, input_filenames = input_data(FLAGS.captcha_dir)
images = tf.constant(input_images)
logits = captcha.inference(images, keep_prob=1.0)
result = captcha.output(logits)
saver = tf.train.Saver()
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
print(tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
recog_result = sess.run(result)
sess.close()
text = one_hot_to_texts(recog_result)
print('text: ', text)
total_count = len(input_filenames)
true_count = 0
for i in range(total_count):
print('image ' + input_filenames[i] + " recognize ----> '" +
text[i] + "'")
if text[i] in input_filenames[i]:
true_count += 1
precision = true_count / total_count
print('%s true/total: %d/%d recognize @ 1 = %.3f' %
(datetime.now(), true_count, total_count, precision))
