astarlight/cps-ocr-engine

An awesome OCR engine developed by SYSU DeepDriving Lab

cps-ocr-engine's Issues

absl.flags._exceptions.IllegalFlagValueError: flag --batch_size=128: ('Non-boole an argument to boolean flag', 128)

Traceback (most recent call last):
File "Chinese_OCR.py", line 48, in
tf.app.flags.DEFINE_boolean('batch_size', 128, 'Validation batch size')
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/flags. py", line 58, in wrapper
return original_function(*args, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/absl/flags/_defines.py", line 267 , in DEFINE_boolean
DEFINE_flag(_flag.BooleanFlag(name, default, help, **args),
File "/usr/local/lib/python2.7/dist-packages/absl/flags/_flag.py", line 293, i n init
p, None, name, default, help, short_name, 1, **args)
File "/usr/local/lib/python2.7/dist-packages/absl/flags/_flag.py", line 107, i n init
self._set_default(default)
File "/usr/local/lib/python2.7/dist-packages/absl/flags/_flag.py", line 196, i n _set_default
self.default = self._parse(value)
File "/usr/local/lib/python2.7/dist-packages/absl/flags/_flag.py", line 169, i n _parse
'flag --%s=%s: %s' % (self.name, argument, e))
absl.flags._exceptions.IllegalFlagValueError: flag --batch_size=128: ('Non-boole an argument to boolean flag', 128)
请问遇到这个问题如何解决的？

第45行的tf.app.flags.DEFINE_boolean('batch_size', 128, 'Validation batch size')

第45行的tf.app.flags.DEFINE_boolean('batch_size', 128, 'Validation batch size')，在OS，python3.6,tensorflow1.4中返回错误，必需是一个BOOL量
另外想请教一下博主，在输出中，ACCURACY始终是0.0，但是LOSS数值在减少，这个正常吗？
谢谢

请问算法可视化这个是用什么做的？前端？QT？

如题

数据在哪里呢？

请问数据在哪里下载呢？没有对应的文件夹 datasets/train

训练成功，但是识别错误

python2.7 显卡是GTX960
1、训练成功了，但是识别tmp，目录的文字会是失败，（如下）不管什么字，都识别成一字？？？ 重新训练一次，识别的结果会不同，但是所有字都识别成了一个字的错误仍在
[the result info] image: ./tmp/0000.jpg predict: 一 丁 七; predict index [[0 1 2]] predict_val [[nan nan nan]]
[the result info] image: ./tmp/0001.jpg predict: 一 丁 七; predict index [[0 1 2]] predict_val [[nan nan nan]]
[the result info] image: ./tmp/0002.jpg predict: 一 丁 七; predict index [[0 1 2]] predict_val [[nan nan nan]]
2、使用作者的模型就可以正常识别
3、gpu版本和cpu版本都试过了，一样的问题
4、我有修改过程序的batch，因为原来的128会出错，改成了32

识别效果很差

我用训练好的模型识别下面的字符。结果一个都不对。怎么破:(

这个是基于Tensorflow吗？

这个是基于Tensorflow吗？

你好，想请教一下添加数字和英文会出现无法分类的情况，请问有一个较好的解决思路吗？

如题，谢谢你的开源

字符输出问题

您好，我复现的时候，训练准确率和测试都没问题，只是最后输出是一些汉字编码，不是直接输出汉字，请问这个怎么解决？

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value fc2/weights [[node fc2/weights/read (defined at Chinese_OCR.py:133) ]] [[node TopKV2 (defined at Chinese_OCR.py:153) ]]

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value fc2/weights
[[node fc2/weights/read (defined at Chinese_OCR.py:133) ]]
[[node TopKV2 (defined at Chinese_OCR.py:153) ]]

help me, thanks

a typo in gen_printed_char.py line 394

in file gen_printed_char.py line 394

if rotate < 0:
        roate = - rotate

should be

if rotate < 0:
        rotate = - rotate

你们没遇到这个错误吗？

TypeError: Using a tf.Tensor as a Python bool is not allowed. Use if t is not None: instead of if t: to test if a tensor is defined, and use TensorFlow ops such as tf.cond to execute subgraphs conditioned on the value of a tensor.

动图里面识别那个900.00太厉害

就算有红章的干扰也能识别全对，是根据通道把红色的部分先去掉了吗

你好，我想生成白色背景，黑色字体的图像，请问代码如何进行修改呢？我目前修改了背景的代码为white：img = Image.new("RGB", (self.width, self.height), "white")，字体为黑：draw.text((0, 0), char, (0, 0, 0), font=font)。但是生成的图像有问题，请问代码要怎么修改呢

训练过程中 accuracy始终不变？

the step 5799.0 takes 8.73126506805 loss 0.443596959114
the step 5800.0 takes 8.83500409126 loss 0.442320615053
the step 5801.0 takes 8.59880495071 loss 0.410989582539
===============Eval a batch=======================
the step 5801.0 test accuracy: 0.0
===============Eval a batch=======================
the step 5802.0 takes 8.64341807365 loss 0.505015850067
the step 5803.0 takes 8.5961420536 loss 0.405117839575
the step 5804.0 takes 8.67368292809 loss 0.427048921585

我训练了5800+次了，我的损失函数和readme中的曲线一致，但是怎么我的accuracy一致都是0.0 啊？不是应该不断上升么？

如果是中英文数字混合的文字，怎么处理数字和英文识别呢？

• 想请教下楼主，如果是中英文数字混合的文字，怎么处理数字和英文识别呢？

有问题想请教您

有问题想请教您,我导师给我的毕业论文方向跟您做的很接近,但是我不懂,您可以指导我一下吗?毕业不了,很烦

tensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value conv3_5/BatchNorm/moving_mean [[Node: conv3_5/BatchNorm/moving_mean/read = Identity[T=DT_FLOAT, _class=["loc:@conv3_5/BatchNorm/moving_mean"], _device="/job:localhost/replica:0/task:0/gpu:0"](conv3_5/BatchNorm/moving_mean)]]

直接测试报了这样一个错误，大家有没有遇到
FailedPreconditionError (see above for traceback): Attempting to use uninitialized value conv3_5/BatchNorm/moving_mean
[[Node: conv3_5/BatchNorm/moving_mean/read = IdentityT=DT_FLOAT, _class=["loc:@conv3_5/BatchNorm/moving_mean"], _device="/job:localhost/replica:0/task:0/gpu:0"]]

请问当我添加噪点以后，start training 后就结束了是什么原因

谢谢

只训练黑体字怎么办

楼主你好，我现在只针对黑体汉字做训练，是不是只需保留font里面的heiti.ftt就可以了，需要修改label文件吗，如果没有gpu，是否可以用cpu训练呢，请您给我一些建议

label_dict = pickle.load(f) 异常退出

def get_label_dict():
f = open('./chinese_labels', 'r')
label_dict = pickle.load(f)
# label_dict = str.encode(pickle.load(f))
f.close()
return label_dict

提示下面的错误：
Traceback (most recent call last):
File "F:/OCR/CPS-OCR-Engine-master/ocr/Chinese_OCR.py", line 411, in
tf.app.run()
File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\platform\app.py", line 125, in run
_sys.exit(main(argv))
File "F:/OCR/CPS-OCR-Engine-master/ocr/Chinese_OCR.py", line 390, in main
label_dict = get_label_dict()
File "F:/OCR/CPS-OCR-Engine-master/ocr/Chinese_OCR.py", line 335, in get_label_dict
label_dict = pickle.load(f)
TypeError: a bytes-like object is required, not 'str'

这个该怎么改？

How to use or port to android?

Hi,
现在各种OCR识别都是在云端的，Tess倒是可以本地识别，但太烂，您的这个有没有办法一直到Android 端，本地识别是因为，要在无网络的环境中使用，不得已如此

运行生成字符数据时的问题

Traceback (most recent call last):
File "gen_printed_char.py", line 380, in
label_dict = get_label_dict()
File "gen_printed_char.py", line 304, in get_label_dict
label_dict = pickle.load(f)
TypeError: a bytes-like object is required, not 'str'

win10和Ubantu都报这个错，不知道为什么

想问下数十类票据分类的大致思路是什么样的啊

想问下数十类票据分类的大致思路是什么样的啊？

有没有可能用深度学习的方法做到自动分类？不用去配置题头区域啥的，我看百度的自定义模板是用至少四个参照区域 http://ai.baidu.com/forum/topic/show/497372

python3.5.2 node fc2/BatchNorm/moving_mean/read 怎么办！？

Caused by op 'fc2/BatchNorm/moving_mean/read', defined at:
File "Chinese_OCR.py", line 460, in
tf.app.run()
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 125, in run
_sys.exit(main(argv))
File "Chinese_OCR.py", line 444, in main
final_predict_val, final_predict_index = inference(name_list)
File "Chinese_OCR.py", line 406, in inference
graph = build_graph(top_k=3)
File "Chinese_OCR.py", line 159, in build_graph
scope='fc2')
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 1866, in fully_connected
outputs = normalizer_fn(outputs, **normalizer_params)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 597, in batch_norm
scope=scope)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 354, in _fused_batch_norm
collections=moving_mean_collections)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/variables.py", line 350, in model_variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/variables.py", line 277, in variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 1479, in get_variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 1220, in get_variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 530, in get_variable
return custom_getter(**custom_getter_kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 1750, in layer_variable_getter
return _model_variable_getter(getter, *args, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 1741, in _model_variable_getter
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/variables.py", line 350, in model_variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 182, in func_with_args
return func(*args, **current_args)
File "/usr/local/lib/python3.5/site-packages/tensorflow/contrib/framework/python/ops/variables.py", line 277, in variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 499, in _true_getter
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 911, in _get_single_variable
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 213, in call
return cls._variable_v1_call(*args, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 176, in _variable_v1_call
aggregation=aggregation)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 155, in
previous_getter = lambda **kwargs: default_variable_creator(None, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variable_scope.py", line 2495, in default_variable_creator
expected_shape=expected_shape, import_scope=import_scope)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 217, in call
return super(VariableMetaclass, cls).call(*args, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 1395, in init
constraint=constraint)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/variables.py", line 1557, in _init_from_args
self._snapshot = array_ops.identity(self._variable, name="read")
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/util/dispatch.py", line 180, in wrapper
return target(*args, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/array_ops.py", line 81, in identity
ret = gen_array_ops.identity(input, name=name)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/ops/gen_array_ops.py", line 3890, in identity
"Identity", input=input, name=name)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/framework/op_def_library.py", line 788, in _apply_op_helper
op_def=op_def)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func
return func(*args, **kwargs)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 3300, in create_op
op_def=op_def)
File "/usr/local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 1801, in init
self._traceback = tf_stack.extract_stack()

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value fc2/BatchNorm/moving_mean
[[node fc2/BatchNorm/moving_mean/read (defined at Chinese_OCR.py:159) ]]

能否分享一下做单字切割的代码？

An issue while training with low resolution image

Dear AstarLight:
I am using some low resolution image(13*13) to train the Engine. I am quite confused about why and which of the following way is better.

1. Using low resolution images as input to train the model directly. Right now, I am using this way. However, the accuracy of the model keep in a quit low level.
2. resize the input image shape to 64*64, then train the model.
3. just used the existed model to evaluate my low resolution images
   Or any other suggestion?

运行train时卡主的问题。

/Applications/anaconda3/envs/tensorflow/bin/python3 /Users/fsl/CPS-OCR-Engine/ocr/Chinese_OCR.py --mode=train --max_steps=16002 --eval_steps=100 --save_steps=500
2019-01-25 15:21:59.993717: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
WARNING:tensorflow:From /Users/fsl/CPS-OCR-Engine/ocr/Chinese_OCR.py:91: slice_input_producer (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by tf.data. Use tf.data.Dataset.from_tensor_slices(tuple(tensor_list)).shuffle(tf.shape(input_tensor, out_type=tf.int64)[0]).repeat(num_epochs). If shuffle=False, omit the .shuffle(...).
train
Begin training
./dataset/train/03755
./dataset/test/03755
WARNING:tensorflow:From /Applications/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/input.py:372: range_input_producer (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by tf.data. Use tf.data.Dataset.range(limit).shuffle(limit).repeat(num_epochs). If shuffle=False, omit the .shuffle(...).
WARNING:tensorflow:From /Applications/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/input.py:318: input_producer (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by tf.data. Use tf.data.Dataset.from_tensor_slices(input_tensor).shuffle(tf.shape(input_tensor, out_type=tf.int64)[0]).repeat(num_epochs). If shuffle=False, omit the .shuffle(...).
WARNING:tensorflow:From /Applications/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/input.py:188: limit_epochs (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by tf.data. Use tf.data.Dataset.from_tensors(tensor).repeat(num_epochs).
WARNING:tensorflow:From /Applications/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/input.py:197: QueueRunner.init (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the tf.data module.
WARNING:tensorflow:From /Applications/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/training/input.py:197: add_queue_runner (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the tf.data module.
WARNING:tensorflow:From /Users/fsl/CPS-OCR-Engine/ocr/Chinese_OCR.py:101: shuffle_batch (from tensorflow.python.training.input) is deprecated and will be removed in a future version.
Instructions for updating:
Queue-based input pipelines have been replaced by tf.data. Use tf.data.Dataset.shuffle(min_after_dequeue).batch(batch_size).
WARNING:tensorflow:From /Users/fsl/CPS-OCR-Engine/ocr/Chinese_OCR.py:187: start_queue_runners (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the tf.data module.
:::Training Start:::

之后就一直卡在这里，跑了一个下午都没往下，这是什么问题呢？谢谢。

还有就是看到您的算法可视化中只取了几个区域对票据进行识别，这个是怎么做到的

请问这是没有用gpu训练吗

你好，请问一下，目前的模型能识别发票信息吗？谢谢

ImageDraw

为什么一直出现 no module named "PIL" 我已经安装过了 PIL库 脚本里没报错，但是cmd时出错

请问一下，英文字母和阿拉伯数字的标签怎么添加？

在chinese_labels中怎么添加英文字母和阿拉伯数字标签，有没有大佬给个方案。

IllegalFlagValueError: flag --batch_size=128: ('Non-boolean argument to boolean flag', 128)

When try to trian, got this error:
IllegalFlagValueError: flag --batch_size=128: ('Non-boolean argument to boolean flag', 128)

配置文件

usage: gen_printed_char.py [-h] ---out_dir OUT_DIR --font_dir FONT_DIR
[--test_ratio TEST_RATIO] --width 30 --height 30
[--no_crop] [--margin MARGIN] [--rotate ROTATE]
[--rotate_step ROTATE_STEP] [--need_aug]
gen_printed_char.py: error: the following arguments are required: ---out_dir, --font_dir, --width, --height
出现了这个错误，大家都是改动了哪里的代码啊

运行 gen_printed_char.py 时报错？

3754 龟
一 0
Traceback (most recent call last):
File "gen_printed_char.py", line 424, in
image = font2image.do(verified_font_path, char, rotate=k)
File "gen_printed_char.py", line 295, in do
np_img)
File "gen_printed_char.py", line 210, in do
ret_img = self.put_img_into_center(norm_img, resized_cv2_img)
File "gen_printed_char.py", line 165, in put_img_into_center
start_width:start_width + width_small] = img_small
TypeError: slice indices must be integers or None or have an index method

关于chinese_labels

请问chinese_labels是怎么生成的呢？

希望作者能提供一下环境搭建的指南，谢谢

只看出作者使用的是py2，所以麻烦作者能告知如何搭建环境，谢谢

训练过程中 accuracy始终不变？

还未解决，有人解决了吗

please change your example about the train_data generating

the code about generating the data_set is not the same as the size in your training code,
python gen_printed_char.py --out_dir ./dataset --font_dir ./chinese_fonts --width 30 --height 30 --margin 4 --rotate 30 --rotate_step 1
so,the size of 30*30 will get an error in the next step of training the model.
the right way to get the true model is to change the code into
python gen_printed_char.py --out_dir ./dataset --font_dir ./chinese_fonts --width 64 --height 64 --margin 4 --rotate 30 --rotate_step 1

代码是不是不全呀

这个Chinese_OCR.py 里面的第20行reload（sys）报错

汉字用30x30的分辨率是不是有点低？

我看gen_printed_char 产生的都是30x30的分辨率，对有些复杂的汉字，这个分辨率是不是有点低呢？后面又resize为64x64的图片，我能不能直接 gen_printed_char 产生出 64x64的图片呢？网络不需要修改吧？

训练数据的前景色和背景色对训练效率和识别准确度有影响吗？

想请问下为什么特意把训练数据设置成黑底白字？
感觉白底黑字的话因为array大部分都是0是不是对训练效率有提升呢？对识别准确度有影响吗？

关于字符切割的处理

代码的执行结果与Readme中的结果并不一致。
请问作者在遇到非联通字体和标点符号时是如何处理的？

chinese_labels文件是如何生成的？

请问作者，chinese_labels文件是如何生成的？打开该文件都是一堆数字，并没有看到汉字，不知道如何生成该文件，还请指导！！感谢感谢！

跑完inference 没有切字

你好 我用的cpu版 跑完了没有切字。

Accuracy问题

有好多人提到accuracy不变的问题，我也遇到了这个问题，accuracy=0，我用的是window系统。编码格式的问题我查过了，没问题，当我调整分类的类别数时，发现accuracy不再是0，分类数为2时，accuracy在0.5附近，分类数为10时，accuracy在0.1附近，这说明训练集或测试集用的是一个标签，但我又仔细检查标签后发现标签也没问题，最终没有找出原因，不知能否麻烦您帮我分析一下到底是什么原因导致的accuracy出错，谢谢

您好，麻烦能提供一下3755个（一级字库）

对图像的像素点求和时，不需要全部遍历所有像素点的值，只需要找到一个像素点大于2就行

ValueError: invalid literal for int() with base 10: '.DS_Store'

Traceback (most recent call last):
File "Chinese_OCR.py", line 411, in
tf.app.run()
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 125, in run
_sys.exit(main(argv))
File "Chinese_OCR.py", line 381, in main
train()
File "Chinese_OCR.py", line 176, in train
train_feeder = DataIterator(data_dir='./dataset/train/')
File "Chinese_OCR.py", line 69, in init
self.labels = [int(file_name[len(data_dir):].split(os.sep)[0]) for file_name in self.image_names]
ValueError: invalid literal for int() with base 10: '.DS_Store'