I have got an issue when run the captcha_records.py, the problem is that my captcha length is 6 or 7 characters, so I set CHARS_NUM = 7, when run the captcha_records.py, it shows an error
"IndexError: index 7 is out of bounds for axis 0 with size 7"
How can I apply for dynamic length of captcha?

Thank you very much!

your code helped me a lot !
so appreciate your fantastic work...
can i send email to you and ask some question about the code in this repository?

I used Tensorflow-CPU. The captcha_train.py had run over 3 hours and the step was over 2700, but the loss was between 2.9 and 3. It decreased very slowly. Can you tell me how long did your training.py run? And, doed it stop until the loss reach zero ? THX

我想用golang去获取训练结果，识别验证码。
刚接触，不太懂。

我在用你的程序做验证码识别，每次识别都会重新加载模型，有什么办法只让模型加载一次吗?

Copying the exact error while running captcha_records.py

(tensorenviron) C:\Work\captcha_recognize-master>python captcha_records.py
Looking for images in './data/train_data'

Writing ./data\train.tfrecords
Traceback (most recent call last):
File "captcha_records.py", line 123, in
tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
File "C:\Users\User\AppData\Local\Continuum\anaconda2\envs\tensorenviron\lib\site-packages\tensorflow\python\platform\app.py", line 125, in run
_sys.exit(main(argv))
File "captcha_records.py", line 102, in main
conver_to_tfrecords(training_data, TRAIN_FILE)
File "captcha_records.py", line 55, in conver_to_tfrecords
num_examples = len(data_set)
TypeError: object of type 'zip' has no len()

Hi.
in captcha_records.py stroke 123 it says:
AttributeError: module 'tensorflow' has no attribute 'app'
i changed tf.app.run(main=main, argv=[sys.argv[0]] + unparsed) to tf.compat.v1.run(main=main, argv=[sys.argv[0]] + unparsed)
now it says module 'tensorflow.compat.v1' has no attribute 'run'

框架能否可以在暂停后再次按原有的进度继续训练，甚者能否支持模型保存后的后续优化训练？

Model doesnot work. Training loss never stablizes. Here is the proof.

Step 440 run_train: loss = 15.94 (8.119 sec)
Step 450 run_train: loss = 36.65 (8.486 sec)
Step 460 run_train: loss = 102.19 (8.480 sec)
Step 470 run_train: loss = 317.27 (9.262 sec)
Step 480 run_train: loss = 801.44 (8.208 sec)
Step 490 run_train: loss = 2089.72 (7.839 sec)
Step 500 run_train: loss = 5845.62 (7.887 sec)
2017-08-26 13:12:49.057376 Saving in ./captcha_train/captcha
Step 510 run_train: loss = 13428.68 (7.818 sec)
Step 520 run_train: loss = 25587.19 (9.801 sec)
Step 530 run_train: loss = 59853.98 (7.978 sec)
Step 540 run_train: loss = 108795.05 (8.611 sec)
Step 550 run_train: loss = 190373.88 (8.281 sec)
Step 560 run_train: loss = 308137.38 (7.726 sec)

Configuration:
IMAGE_HEIGHT = 65
IMAGE_WIDTH = 190
CHAR_SETS = 'abcdefghijklmnopqrstuvwxyz'
CLASSES_NUM = len(CHAR_SETS)
CHARS_NUM = 8

I tried freezing the checkpoints with "Reshape_1:0" as an output Tensor, but after converting into ProtoBuffer I am unable to generate an inference from it.

What should be right input and output nodes for the Frozen graph?

When training your model, you use CPU, GPU or multi GPUs? If GPU, what's the version and how long did you take about training your model?

I get this

$python ./captcha_gen_default.py 
>> generate 1000 captchas in ./data/test_data
Traceback (most recent call last):
  File "./captcha_gen_default.py", line 55, in <module>
    gen(FLAGS.test_dir, TEST_SIZE, CHARS_NUM)
  File "./captcha_gen_default.py", line 28, in gen
    for i in xrange(total_size):
NameError: name 'xrange' is not defined

what I can to do?

Writing ./data/train.tfrecords
Traceback (most recent call last):
File "captcha_records.py", line 123, in
tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
File "~/.local/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "captcha_records.py", line 102, in main
conver_to_tfrecords(training_data, TRAIN_FILE)
File "captcha_records.py", line 62, in conver_to_tfrecords
label_raw = label_to_one_hot(label).tostring()
File "captcha_records.py", line 42, in label_to_one_hot
index.append(CHAR_SETS.index(c))
ValueError: substring not found

Before adding _1 to all of my training captchas I've got this error:

Traceback (most recent call last):
File "captcha_records.py", line 123, in
tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)
File "~/.local/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "captcha_records.py", line 102, in main
conver_to_tfrecords(training_data, TRAIN_FILE)
File "captcha_records.py", line 62, in conver_to_tfrecords
label_raw = label_to_one_hot(label).tostring()
File "captcha_records.py", line 44, in label_to_one_hot
one_hot_label[one_hot_index] = 1.0
IndexError: index 5 is out of bounds for axis 0 with size 5

hi
can you update your python versionto 3.6 ?

extensions = ['jpg', 'JPG', 'jpeg', 'JPEG', 'png', 'PNG']
# ...
for extension in extensions:
     file_glob = os.path.join(image_dir, '*.' + extension)
     file_list.extend(gfile.Glob(file_glob))

在Windows上，后缀大小写不区分，同一张图片会被载入两次：

请问下换成其他的图形验证码尺寸需要怎么调整。。谢谢！！

Hi everybody!
I've followed all the readme file.
But: how I can "scan" my image?

Is it possible to make it recognize multiple image sizes different from the training ones? Im having tensor shapes different error while trying with images bigger than train ones.

您好！感谢您的分享，请问关于这个项目您有博客文章或者任何其他形式的理论说明吗？

请问我若过用一个cpu来训练需要多长时间，
IMAGE_HEIGHT = 60
IMAGE_WIDTH = 150
CHARS_NUM = 4
准确率能达到多少，因为我想尽快用到，所以比较急，谢谢

Does it work with Multi Threading? I tried, but received some errors.

On https://github.com/aero2a/kape-
is file https://github.com/aero2a/kape-/blob/master/dane.txt
'name of file.jpg' 'result in txt' not important

is possible to solve this? I foudn file to generate (in php https://github.com/letarg0/obrazy/blob/master/my.php)

I am not able to convert to tfjs model(webmodel) using tensorflowjs_converter
throws the Error msg:
File "/home/shiva/.local/bin/tensorflowjs_converter", line 11, in
sys.exit(main())
File "/home/shiva/.local/lib/python2.7/site-packages/tensorflowjs/converters/converter.py", line 373, in main
(input_format, output_format))
ValueError: Unsupported input_format - output_format pair: tf_saved_model - tfjs_layers_model

Looking for images in './data/test_data'
2018-04-16 15:11:21.166594: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.2
./captcha_train/captcha-62400
terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted

Based on the code, it seems the main method in the captcha_train.py deletes recursively the training data. For neural networks that take longer to train, it is useful to be able to fine tune the network along the way. Unless I am interpreting the code wrong?

what's the theory? how can you recognize the whole captcha without image segmentation ?

the loss is reduce to 0 when step 170 , but it don't stop,
i pressed 'ctrl+c' breaked it, then try to eval and recognize,
but the result of recognize is so bad,

can i break the train when loss equal 0?
or it will finished , i need't do that?