kasyoukin/caffe_ocr_for_linux

caffe_ocr_for_linux's Issues

编译成功 生成ocr_test最后出现ocr_test imgfolder modelfolder [chi|eng]

请问这是什么意思

develop warp-ctc

你好，develop warp-ctc安装报错：

你用的这个 develop warp-ctc能编译通过？有没有用GPU啊？
方便的话，加个联系方式，讨教一下，我邮箱：[email protected]

ctcpp.h

感谢分享，我在安装完warp-ctc之后发现/usr/local/include/下有ctc.h，但源码中并没有发现ctcpp.h，请问ctcpp.h在哪里呢？

undefined reference to `compute_ctc_loss'

您好，我进行编译的时候出现了如下错误：
build_release/lib/libcaffe.so: undefined reference to ctcStatus_t CTC::get_workspace_size<float>(int const*, int const*, int, int, ctcOptions, unsigned long*)' .build_release/lib/libcaffe.so: undefined reference to compute_ctc_loss'
.build_release/lib/libcaffe.so: undefined reference to ctcStatus_t CTC::compute_ctc_loss<double>(double const*, double*, int const*, int const*, int const*, int, int, double*, void*, ctcOptions)' .build_release/lib/libcaffe.so: undefined reference to ctcStatus_t CTC::get_workspace_size(int const*, int const*, int, int, ctcOptions, unsigned long*)'
.build_release/lib/libcaffe.so: undefined reference to `ctcGetStatusString'
collect2: error: ld returned 1 exit status
这个是否由于我的warp-ctc没有装好？我将warp-ctc编译完之后，除了修改makefile.config的路径之外，还有将ctcpp.h以及libwarpctc.so拷贝到/usr/lib下，是否还有其他遗漏步骤？
万分感谢！

ctcpp.h

没有找到ctcpp.h
https://github.com/ChWick/warp-ctc 这个里面也是只有ctc.h
这两个不一样吗？能否提供一个ctcpp.h

已解决，在https://github.com/ChWick/warp-ctc分支中找到

关于训练出现ctcloss=0怎么解决

I0926 10:52:45.730175 5561 solver.cpp:237] Train net output #0: ctcloss = 0 (* 1 = 0 loss)
I0926 10:52:45.730180 5561 sgd_solver.cpp:105] Iteration 3000, lr = 0.01
I0926 10:52:47.260527 5572 data_layer.cpp:78] Restarting data prefetching from start.
^CI0926 10:52:47.829211 5561 solver.cpp:447] Snapshotting to binary proto file densenet-res-blstm_train-val_iter_3104.caffemodel

protobuf version?

编译错误

你好，编译必须要用到opencv吗，报以下错误

AR -o .build_release/lib/libcaffe.a
LD -o .build_release/lib/libcaffe.so.1.0.0
/usr/bin/ld: cannot find -lopencv_imgcodecs
collect2: error: ld returned 1 exit status
Makefile:581: recipe for target '.build_release/lib/libcaffe.so.1.0.0' failed
make: *** [.build_release/lib/libcaffe.so.1.0.0] Error 1

F0725 12:34:21.153914 43670 db.cpp:20] Unknown database backend

这是怎么回事，最后一步caffe train 的时候出现错误了

Makefile:601: recipe for target '.build_release/cuda/src/caffe/layers/DenseBlock_layer.o' failed

I ran into the following error:

NVCC src/caffe/layers/cudnn_sigmoid_layer.cu
/usr/local/include/boost/core/noncopyable.hpp(42): error: defaulted default constructor cannot be constexpr because the corresponding implicitly declared default constructor would not be constexpr

1 error detected in the compilation of "/tmp/tmpxft_00006a5e_00000000-11_DenseBlock_layer.compute_61.cpp1.ii".
Makefile:601: recipe for target '.build_release/cuda/src/caffe/layers/DenseBlock_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/DenseBlock_layer.o] Error 1
make: *** Waiting for unfinished jobs....

what should I do?
Thanks.

只能识别一个汉字

这是因为虽然做了格式转换ANTI to UTF8 但是没有做dos2unix label.txt，导致机器不能识别label中的换行，所以只能读取一个字符
如果在docker 下不能运行dos2unix命令，可以切换出去执行完dos2unxi命令之后再切换回去
最后执行的效果是
[1/1]/ai/zhaoliang/7_caffe_ocr_for_linux/caffe_ocr_for_linux/datatest/20438234_4066860539.jpg[cost:18.098 ms]: 宛转万态，总的来说，

cuDNN卷积模式修改导致问题

你好！

我是NVIDIA软件研发部的Nic，我开始时候用你的代码训练CRNN模型没有问题。
后来将denseblocklayer.cu第382行的卷积模式改成了cross-corralation，重新训练了模型后识别效果很差。
不知道为什么。。
请问你原来为什么不用cross-corralation模式？Caffe默认只支持cross-corralation模式。
Thanks.

/convert_mnist_siamese_data.cpp:96:24: error

examples/siamese/convert_mnist_siamese_data.cpp:96:24: error: no matching function for call to ‘caffe::Datum::set_label(int)’
datum.set_label(1);
^
In file included from examples/siamese/convert_mnist_siamese_data.cpp:15:0:
.build_release/src/caffe/proto/caffe.pb.h:11901:13: note: candidate: void caffe::Datum::set_label(int, google::protobuf::int32)
inline void Datum::set_label(int index, ::google::protobuf::int32 value) {
^
.build_release/src/caffe/proto/caffe.pb.h:11901:13: note: candidate expects 2 arguments, 1 provided
examples/siamese/convert_mnist_siamese_data.cpp:98:24: error: no matching function for call to ‘caffe::Datum::set_label(int)’
datum.set_label(0);
^
In file included from examples/siamese/convert_mnist_siamese_data.cpp:15:0:
.build_release/src/caffe/proto/caffe.pb.h:11901:13: note: candidate: void caffe::Datum::set_label(int, google::protobuf::int32)
inline void Datum::set_label(int index, ::google::protobuf::int32 value) {
^
.build_release/src/caffe/proto/caffe.pb.h:11901:13: note: candidate expects 2 arguments, 1 provided
Makefile:588: recipe for target '.build_release/examples/siamese/convert_mnist_siamese_data.o' failed
make: *** [.build_release/examples/siamese/convert_mnist_siamese_data.o] Error 1
大家有没有遇到这个问题，谢谢

输出的长度特别长，，有很多0是怎么回事

训练的时候label的长度是5，设置image size 是128*256，输出的时候Label长度有好几十，且大部分是0

▒▒/1]/home/zhangtao/work/caffe_ocr_for_linux/data/img/20438234_4066860539.jpg[cost:20.3835 ms]: ▒▒

我用inception-bn-res-blstm chi 测试出来是这
转码报错
zhangtao@BoHong:/work/caffe_ocr_for_linux/models/inception-bn-res-blstm$ iconv -f GBK -t UTF-8 label.txt > mylabel.txt
iconv: illegal input sequence at position 16333
zhangtao@BoHong:/work/caffe_ocr_for_linux/models/inception-bn-res-blstm$ iconv -f ANSI -t UTF-8 label.txt > mylabel.txt
iconv: conversion from ANSI' is not supported Try iconv --help' or `iconv --usage' for more information.

编译和测试没问题，但在训练模型时出现warp_ctc_loss_layer.cu:118] Check failed: status == CTC_STATUS_SUCCESS (3 vs. 0)

你好，编译通过了，测试也没有问题，但是在训练模型时出现warp_ctc_loss_layer.cu:118] Check failed: status == CTC_STATUS_SUCCESS (3 vs. 0) CTC Error: execution failed
怎么解决呢？谢谢

ocr_test.cpp编译问题

感谢分享！

在编译caffe的过程中遇到这个问题

ctc.h和ctcpp.h在　/usr/local/include路径中

Makefile.config

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include

编译错误信息

ocr_test.cpp:(.text+0x265)：对‘ctcStatus_t CTC::get_workspace_size<float>(int const*, int const*, int, int, ctcOptions, unsigned long*)’未定义的引用
ocr_test.cpp:(.text+0x44f)：对‘ctcStatus_t CTC::compute_ctc_loss<float>(float const*, float*, int const*, int const*, int const*, int, int, float*, void*, ctcOptions)’未定义的引用

不知道哪里错了，寻求一下您的帮助，谢谢．

无

选择哪块显卡运行ocr_test

我在进行ocr_test测试时，使用第0块显卡；现在我想通过命令行参数传递使用哪块选卡，于是做了如下修改
bool Classifier::Init(const string& model_path, int gpu_id)
{
const string trained_file = model_path + "/model.caffemodel";
const string model_file = model_path + "/deploy.prototxt";
string mean_file = model_path + "/mean.binaryproto";
const string mean_value_file = model_path + "/mean_values.txt";
const string label_file = model_path + "/label.txt";

if (!CheckFileExist(mean_file.c_str()))
	mean_file = mean_value_file;
if (-1 == gpu_id)
{
    LOG(INFO) << "Use CPU.";
    Caffe::set_mode(Caffe::CPU);
}
else
{
    LOG(INFO) << "Use GPU with device ID " << gpu_id;
    cudaDeviceProp device_prop;
    cudaGetDeviceProperties(&device_prop, gpu_id);
    LOG(INFO) << "GPU device name: " << device_prop.name;
	Caffe::SetDevice(gpu_id);
    Caffe::set_mode(Caffe::GPU);
}

重新编译运行时，选择第3块显卡，结果第0块和第三块显卡一起在跑，请问什么原因啊
但是我跑这种./build/tools/caffe test却是正常的

乱码

??/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20436765_2556130357.jpg[cost:82.9587 ms]: 3
??/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20436328_800384098.jpg[cost:4.41216 ms]: ?
???/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20440484_2127180630.jpg[cost:4.37731 ms]: ?
??/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20437703_937698549.jpg[cost:4.3505 ms]: ?
???/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20438234_4066860539.jpg[cost:4.32771 ms]: ?
??/6]/home/whsyst/caffe_ocr_for_linux/models/ocr/dataset/20436312_1683447152.jpg[cost:4.33046 ms]: ?

Unknown layer type:

cannot find -lwrapctc

我之前在一台电脑上跑通了，换了一台电脑就。。。

编译错误

你好，我在安装好warp-ctc后，编译时出现如下错误，想请教一下是ctc在与cuda的联结上有什么问题吗？

LMDB支持

Makefile.config中，打开USE_LMDB后编译就会报错。

cannot find lopenblas

can you tell me about this question that I have meet

如何创建训练数据？报错LMDB格式不对。。。

验证集上准确率很高，测试集上准确率很低是怎么回事

测试程序最终会编译在./build/tools/ocr_test是什么意思？

Test Net output 输出的有两个acc,分别指的是什么？

就像这样

I0322 17:23:39.161901 17215 solver.cpp:330] Iteration 2500, Testing net (#0)
I0322 17:24:01.905066 17215 solver.cpp:397]     Test net output #0: acc = 0.773306
I0322 17:24:01.905092 17215 solver.cpp:397]     Test net output #1: acc = 0.0848667

训练时将训练样本遍历一遍后，再shuffle一下样本顺序

这个操作没有啊

把ocr_test改造下编译成.so怎么做？

想问下要想把ocr_test改造下编译成.so，makefile应该怎么改啊，谢谢啊

DenseBlock编译错误

In file included from src/caffe/layers/DenseBlock_layer.cpp:19:0:
./include/caffe/layers/DenseBlock_layer.hpp:147:3: error: ‘cudnnTensorDescriptor_t’ does not name a type
cudnnTensorDescriptor_t * quadG_tensorDesc;
^
./include/caffe/layers/DenseBlock_layer.hpp:148:3: error: ‘cudnnTensorDescriptor_t’ does not name a type
cudnnTensorDescriptor_t * quadG_paramDesc;
^
./include/caffe/layers/DenseBlock_layer.hpp:149:3: error: ‘cudnnConvolutionDescriptor_t’ does not name a type
cudnnConvolutionDescriptor_t* convBC_Descriptor;
^
./include/caffe/layers/DenseBlock_layer.hpp:154:11: error: ‘cudnnFilterDescriptor_t’ was not declared in this scope
vector<cudnnFilterDescriptor_t *> BC_filterDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:154:36: error: template argument 1 is invalid
vector<cudnnFilterDescriptor_t *> BC_filterDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:154:36: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:156:10: error: ‘cudnnConvolutionFwdAlgo_t’ was not declared in this scope
vector<cudnnConvolutionFwdAlgo_t *> conv_FwdAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:156:37: error: template argument 1 is invalid
vector<cudnnConvolutionFwdAlgo_t *> conv_FwdAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:156:37: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:157:10: error: ‘cudnnConvolutionFwdAlgo_t’ was not declared in this scope
vector<cudnnConvolutionFwdAlgo_t *> BC_FwdAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:157:37: error: template argument 1 is invalid
vector<cudnnConvolutionFwdAlgo_t *> BC_FwdAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:157:37: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:158:10: error: ‘cudnnConvolutionBwdFilterAlgo_t’ was not declared in this scope
vector<cudnnConvolutionBwdFilterAlgo_t *> conv_BwdFilterAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:158:43: error: template argument 1 is invalid
vector<cudnnConvolutionBwdFilterAlgo_t *> conv_BwdFilterAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:158:43: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:159:10: error: ‘cudnnConvolutionBwdFilterAlgo_t’ was not declared in this scope
vector<cudnnConvolutionBwdFilterAlgo_t *> BC_BwdFilterAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:159:43: error: template argument 1 is invalid
vector<cudnnConvolutionBwdFilterAlgo_t *> BC_BwdFilterAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:159:43: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:160:10: error: ‘cudnnConvolutionBwdDataAlgo_t’ was not declared in this scope
vector<cudnnConvolutionBwdDataAlgo_t *> conv_BwdDataAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:160:41: error: template argument 1 is invalid
vector<cudnnConvolutionBwdDataAlgo_t > conv_BwdDataAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:160:41: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:161:10: error: ‘cudnnConvolutionBwdDataAlgo_t’ was not declared in this scope
vector<cudnnConvolutionBwdDataAlgo_t > BC_BwdDataAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:161:41: error: template argument 1 is invalid
vector<cudnnConvolutionBwdDataAlgo_t > BC_BwdDataAlgoVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:161:41: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:172:3: error: ‘cudnnHandle_t’ does not name a type
cudnnHandle_t cudnnHandlePtr;
^
./include/caffe/layers/DenseBlock_layer.hpp:174:10: error: ‘cudnnHandle_t’ was not declared in this scope
vector<cudnnHandle_t> extraHandles;
^
./include/caffe/layers/DenseBlock_layer.hpp:174:24: error: template argument 1 is invalid
vector<cudnnHandle_t> extraHandles;
^
./include/caffe/layers/DenseBlock_layer.hpp:174:24: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:177:10: error: ‘cudnnTensorDescriptor_t’ was not declared in this scope
vector<cudnnTensorDescriptor_t *> tensorDescriptorVec_conv_x;//local Conv X
^
./include/caffe/layers/DenseBlock_layer.hpp:177:35: error: template argument 1 is invalid
vector<cudnnTensorDescriptor_t *> tensorDescriptorVec_conv_x;//local Conv X
^
./include/caffe/layers/DenseBlock_layer.hpp:177:35: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:178:3: error: ‘cudnnTensorDescriptor_t’ does not name a type
cudnnTensorDescriptor_t * tensorDescriptor_conv_y;//local Conv Y
^
./include/caffe/layers/DenseBlock_layer.hpp:179:10: error: ‘cudnnTensorDescriptor_t’ was not declared in this scope
vector<cudnnTensorDescriptor_t *> tensorDescriptor_BN;//
^
./include/caffe/layers/DenseBlock_layer.hpp:179:35: error: template argument 1 is invalid
vector<cudnnTensorDescriptor_t *> tensorDescriptor_BN;//
^
./include/caffe/layers/DenseBlock_layer.hpp:179:35: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:181:10: error: ‘cudnnDropoutDescriptor_t’ was not declared in this scope
vector<cudnnDropoutDescriptor_t *> dropoutDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:181:36: error: template argument 1 is invalid
vector<cudnnDropoutDescriptor_t *> dropoutDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:181:36: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:183:10: error: ‘cudnnFilterDescriptor_t’ was not declared in this scope
vector<cudnnFilterDescriptor_t > filterDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:183:35: error: template argument 1 is invalid
vector<cudnnFilterDescriptor_t > filterDescriptorVec;
^
./include/caffe/layers/DenseBlock_layer.hpp:183:35: error: template argument 2 is invalid
./include/caffe/layers/DenseBlock_layer.hpp:185:3: error: ‘cudnnActivationDescriptor_t’ does not name a type
cudnnActivationDescriptor_t ReLUDesc;
^
./include/caffe/layers/DenseBlock_layer.hpp:187:3: error: ‘cudnnConvolutionDescriptor_t’ does not name a type
cudnnConvolutionDescriptor_t conv_Descriptor;
^
Makefile:588: recipe for target '.build_release/src/caffe/layers/DenseBlock_layer.o' failed
make: *** [.build_release/src/caffe/layers/DenseBlock_layer.o] Error 1
make: *** Waiting for unfinished jobs....

你好，make之后出现这个错误，不知道如何修改

编译报错

### /usr/local/include/boost/system/error_code.hpp:233:21: error: looser throw specifier for ‘virtual const char boost::system::error_category::std_category::name() const’
virtual const char * name() const BOOST_NOEXCEPT
^
/usr/include/c++/5/system_error:77:21: error: overriding ‘virtual const char std::_V2::error_category::name() const noexcept’
name() const noexcept = 0;
^
/usr/local/include/boost/system/error_code.hpp:243:37: error: looser throw specifier for ‘virtual std::error_condition boost::system::error_category::std_category::default_error_condition(int) const’
virtual std::error_condition default_error_condition( int ev ) const
^
/usr/include/c++/5/system_error:104:25: error: overriding ‘virtual std::error_condition std::_V2::error_category::default_error_condition(int) const noexcept’
default_error_condition(int __i) const noexcept;
^
/usr/local/include/boost/system/error_code.hpp:245:21: error: looser throw specifier for ‘virtual bool boost::system::error_category::std_category::equivalent(int, const std::error_condition&) const’
virtual bool equivalent( int code, const std::error_condition & condition ) const
^
/usr/include/c++/5/system_error:107:14: error: overriding ‘virtual bool std::_V2::error_category::equivalent(int, const std::error_condition&) const noexcept’
equivalent(int __i, const error_condition& __cond) const noexcept;
^
/usr/local/include/boost/system/error_code.hpp:247:21: error: looser throw specifier for ‘virtual bool boost::system::error_category::std_category::equivalent(const std::error_code&, int) const’
virtual bool equivalent( const std::error_code & code, int condition ) const
^
/usr/include/c++/5/system_error:110:14: error: overriding ‘virtual bool std::_V2::error_category::equivalent(const std::error_code&, int) const noexcept’
equivalent(const error_code& __code, int __i) const noexcept;
^
Makefile:601: recipe for target '.build_release/cuda/src/caffe/layers/DenseBlock_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/DenseBlock_layer.o] Error 1
make: *** Waiting for unfinished jobs....
tools/ocr_test.cpp: In function ‘void FindAllImages(const char*, std::vector<std::__cxx11::basic_string >&, bool)’:
tools/ocr_test.cpp:174:27: warning: suggest parentheses around assignment used as truth value [-Wparentheses]
while (ent = readdir(pDir))
^
tools/ocr_test.cpp: In function ‘float GetCTCLoss(float*, int, int, int, const string&, const std::map<wchar_t, int>&)’:
tools/ocr_test.cpp:221:14: warning: variable ‘status’ set but not used [-Wunused-but-set-variable]
ctcStatus_t status = CTC::get_workspace_size(&len,
^
tools/ocr_test.cpp: In function ‘void test_ocr_english(const string&, const string&, const string&)’:
tools/ocr_test.cpp:279:11: warning: unused variable ‘c’ [-Wunused-variable]
wchar_t c = 0;
^
tools/ocr_test.cpp:359:95: warning: format ‘%d’ expects argument of type ‘int’, but argument 2 has type ‘size_t {aka long unsigned int}’ [-Wformat=]
printf("[%d/%d]%s\n\torig result: %s\n",i+1,imgs.size(),imgs[i].c_str(), strpredict0.c_str());
^
tools/ocr_test.cpp:359:95: warning: format ‘%d’ expects argument of type ‘int’, but argument 3 has type ‘std::vector<std::__cxx11::basic_string >::size_type {aka long unsigned int}’ [-Wformat=]
tools/ocr_test.cpp:325:6: warning: unused variable ‘nok_lexicon’ [-Wunused-variable]
int nok_lexicon = 0;
^
tools/ocr_test.cpp:326:6: warning: unused variable ‘nok_nolexicon’ [-Wunused-variable]
int nok_nolexicon = 0;
^
tools/ocr_test.cpp: In function ‘void test_ocr_chinese(const string&, const string&)’:
tools/ocr_test.cpp:412:46: warning: format ‘%d’ expects argument of type ‘int’, but argument 2 has type ‘std::vector<std::__cxx11::basic_string >::size_type {aka long unsigned int}’ [-Wformat=]
printf("labels:%d\n",alphabets.size());
^
tools/ocr_test.cpp:422:11: warning: unused variable ‘c’ [-Wunused-variable]
wchar_t c = 0;
^
tools/ocr_test.cpp:466:118: warning: format ‘%d’ expects argument of type ‘int’, but argument 2 has type ‘size_t {aka long unsigned int}’ [-Wformat=]
printf("[%d/%d]%s[cost:%g ms]: %s\n", i + 1, imgs.size(), imgs[i].c_str(),timer.MilliSeconds(), strpredict0.c_str());
^
tools/ocr_test.cpp:466:118: warning: format ‘%d’ expects argument of type ‘int’, but argument 3 has type ‘std::vector<std::__cxx11::basic_string >::size_type {aka long unsigned int}’ [-Wformat=]
tools/ocr_test.cpp:431:6: warning: unused variable ‘nok_lexicon’ [-Wunused-variable]
int nok_lexicon = 0;
^
tools/ocr_test.cpp:432:6: warning: unused variable ‘nok_nolexicon’ [-Wunused-variable]
int nok_nolexicon = 0;
^

---

ocr_test.c 引用了ICNNPredict，请问它定义的函数代码在哪可以找到？

你好，其实想找到这个函数的详细代码，CreatePredictInstance（）
方便的话帮个忙，谢谢哈~

请问具体的编译过程是怎样的

从原作者那下载的中文识别的model要怎么用，需要改哪些文件？您的代码下载下来make之后就可以直接运行了吗，不需要加model了吗？

error: cannot convert ‘const double* const’ to ‘const float*’

Could you help on this issue when I execute make command. thx a lot

CXX src/caffe/layers/tanh_layer.cpp
CXX src/caffe/layers/threshold_layer.cpp
CXX src/caffe/layers/tile_layer.cpp
CXX src/caffe/layers/transpose_layer.cpp
CXX src/caffe/layers/warp_ctc_loss_layer.cpp
src/caffe/layers/warp_ctc_loss_layer.cpp: In instantiation of ‘void caffe::WarpCTCLossLayer::Forward_cpu(const std::vector<caffe::Blob>&, const std::vector<caffe::Blob>&) [with Dtype = double]’:
src/caffe/layers/warp_ctc_loss_layer.cpp:276:1: required from here
src/caffe/layers/warp_ctc_loss_layer.cpp:158:31: error: cannot convert ‘const double* const’ to ‘const float*’ for argument ‘1’ to ‘ctcStatus_t compute_ctc_loss(const float*, float*, const int*, const int*, const int*, int, int, float*, void*, ctcOptions)’
);

迭代两万多次之后loss就卡在30不降了

你好，我们做身份证识别，用你的代码训过，很正常，但是精度还不够高。于是就又生成了200W张身份证地址的图片加到你的训练样本里面，这样差不多就有500W张，经过一些调参，但是迭代两万次之后loss就一直卡在30左右不下降了，我们的样本和label应该是可以对的上的，没错。但是loss就是不降，请问你有什么建议呢？

有关训练

我现在把环境都编译成功了 也测试得出了结果 但是不清楚具体怎么训练 大神能不能具体给出训练步骤 请大神指导

how to set warpctc?

I1213 20:59:11.824182 27507 layer_factory.hpp:77] Creating layer ctcloss
F1213 20:59:11.824225 27507 layer_factory.hpp:81] Check failed: registry.count(type) == 1 (0 vs. 1) Unknown layer type: WarpCTCLoss (known types: AbsVal, Accuracy, ArgMax, BNLL, BatchNorm, BatchReindex, Bias, CTCGreedyDecoder, Concat, ContrastiveLoss, Convolution, Crop, Data, Deconvolution, DenseBlock, Dropout, DummyData, ELU, Eltwise, Embed, EuclideanLoss, Exp, Filter, Flatten, HDF5Data, HDF5Output, HingeLoss, Im2col, ImageData, InfogainLoss, InnerProduct, Input, Interp, LRN, LSTM, LSTMUnit, Log, Lstm, MVN, MemoryData, MultinomialLogisticLoss, PReLU, Parameter, Pooling, Power, RNN, ReLU, Reduction, Reshape, Reverse, ReverseTime, SPP, Scale, Sigmoid, SigmoidCrossEntropyLoss, Silence, Slice, Softmax, SoftmaxWithLoss, SoftmaxWithLossMultiLabel, Split, TanH, Threshold, Tile, Transpose, WindowData)
*** Check failure stack trace: ***
@ 0x7f4028755daa (unknown)
I have built warpctc at ,and makefile.config also set :

Whatever else you find you need goes here.

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial /home//warp-ctc/build /home/css/warp-ctc/include
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial /home//warp-ctc/build

只能识别一个字或者一行，目前版本是否有text detect

我编译成功后test了下，有点迷惑，想请教一下目前的版本是不是只能识别单行的图片？
我输入整张图片基本只能识别一个字
@kasyoukin @moucmou

请教大神，务必回复

我是这样来训练的，首先将训练样本生成对应的标签，然后将标签转化成采用convert_imageset转换成对应lmdb格式的，然后采用densenet-res-blstm_train-val.prototxt开始训练，可是ctcloss=0，刚接触深度学习，不知道如何解决问题

warpctc安装事项。

您好，我在安装warpctc时遇到了问题。master版本中没有ctcpp.h；而develop版中有ctcpp.h，编译时出现如下问题：
ubuntu@ubuntu:~/warp-ctc-develop/build$ make
[ 25%] Building NVCC (Device) object CMakeFiles/warpctc.dir/src/warpctc_generated_reduce.cu.o
[ 50%] Building NVCC (Device) object CMakeFiles/warpctc.dir/src/warpctc_generated_ctcpp_entrypoint.cu.o
/home/ubuntu/warp-ctc-develop/src/ctcpp_entrypoint.cu(1): error: this declaration has no storage class or type specifier

/home/ubuntu/warp-ctc-develop/src/ctcpp_entrypoint.cu(1): error: expected a ";"

2 errors detected in the compilation of "/tmp/tmpxft_0000031b_00000000-19_ctcpp_entrypoint.compute_62.cpp1.ii".
CMake Error at warpctc_generated_ctcpp_entrypoint.cu.o.cmake:266 (message):
Error generating file
/home/ubuntu/warp-ctc-develop/build/CMakeFiles/warpctc.dir/src/./warpctc_generated_ctcpp_entrypoint.cu.o

CMakeFiles/warpctc.dir/build.make:63: recipe for target 'CMakeFiles/warpctc.dir/src/warpctc_generated_ctcpp_entrypoint.cu.o' failed
make[2]: *** [CMakeFiles/warpctc.dir/src/warpctc_generated_ctcpp_entrypoint.cu.o] Error 1
CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/warpctc.dir/all' failed
make[1]: *** [CMakeFiles/warpctc.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

关于caffe make时的错误

你好，我的环境是 ubuntu14.04 cudn7.5 ，cudnn5.1
protobuf2.6.1
在编译过程中出现如下错误，怎么回事呢？能帮我看看问题吗？
我初次接触caffe，安装过程中解决问题比较没经验，麻烦耐心帮我解答一下，谢谢

ocr_test result is wrong

Hello, my result is
经/6]../../examples/ocr/20436328_800384098.jpg[cost:109 ms]: 姻
洲/6]../../examples/ocr/20437703_937698549.jpg[cost:104 ms]: 业
风/6]../../examples/ocr/20436765_2556130357.jpg[cost:104 ms]: 3
持/6]../../examples/ocr/20440484_2127180630.jpg[cost:104 ms]: ?
人/6]../../examples/ocr/20436312_1683447152.jpg[cost:135 ms]: 美
，/6]../../examples/ocr/20438234_4066860539.jpg[cost:109 ms]: 究
I use CPU_ONLY mode,and resnet-res-blstm\inception-bn-res-blstm are all wrong like this.

多值标签制作lmdb问题

请问你是使用create_annoset.py还是create_dataset.py来制作
20456343_4045240981.jpg 89 201 241 178 19 94 19 22 26 656
这种训练的lmdb？

关于测试脚本

我用来训练7位的英文字母，训练的时候准确率达到99.9%以上，但是我就用训练集进行测试，结果只有不到3%完全正确，测出来的结果是6位甚至5位，是测试脚本的问题么？

Check failed: N_ == label_seq->num() (2048 vs. 64)

我在cpu模式下测试你训练好的模型没问题，但是在训练自己的数据是出现这个问题。
I0313 01:33:51.547402 51709 net.cpp:406] ctcloss <- fc1x5990
I0313 01:33:51.547411 51709 net.cpp:406] ctcloss <- label
I0313 01:33:51.547551 51709 net.cpp:380] ctcloss -> ctcloss
F0313 01:33:51.547602 51709 warp_ctc_loss_layer.cpp:47] Check failed: N_ == label_seq->num() (2048 vs. 64)
我是用的是inception_bn网络，能大概看出是哪里出问题了吗？谢谢

加载train_lmdb时出现问题

您好，在开始训练时，我遇到了这种情况：
‘I0515 15:14:44.180166 8095 caffe.cpp:213] CUDNN version: 7004
I0515 15:14:44.180531 8095 caffe.cpp:229] Using GPUs 1
I0515 15:14:44.193114 8095 caffe.cpp:234] GPU 1: GeForce GTX 1080 Ti
I0515 15:14:44.809454 8095 solver.cpp:44] Initializing solver from parameters:
test_iter: 230
test_interval: 1800
base_lr: 0.0001
display: 10
max_iter: 400000
lr_policy: "multistep"
gamma: 0.5
momentum: 0.9
weight_decay: 0.0001
snapshot: 1000
snapshot_prefix: "./caffemodel/densenet"
solver_mode: GPU
device_id: 1
random_seed: 1234
net: "/home/zkgao/caffe_ocr_for_linux/examples/crnn/models/dense/densenet-res-blstm_train-val.prototxt"
train_state {
level: 0
stage: ""
}
stepvalue: 30000
stepvalue: 60000
stepvalue: 90000
stepvalue: 120000
stepvalue: 150000
stepvalue: 180000
I0515 15:14:44.809936 8095 solver.cpp:87] Creating training net from net file: /home/zkgao/caffe_ocr_for_linux/examples/crnn/models/dense/densenet-res-blstm_train-val.prototxt
I0515 15:14:44.810540 8095 upgrade_proto.cpp:77] Attempting to upgrade batch norm layers using deprecated params: /home/zkgao/caffe_ocr_for_linux/examples/crnn/models/dense/densenet-res-blstm_train-val.prototxt
I0515 15:14:44.810571 8095 upgrade_proto.cpp:80] Successfully upgraded batch norm layers using deprecated params.
I0515 15:14:44.811134 8095 net.cpp:294] The NetState phase (0) differed from the phase (1) specified by a rule in layer data
I0515 15:14:44.811242 8095 net.cpp:294] The NetState phase (0) differed from the phase (1) specified by a rule in layer acc
I0515 15:14:44.811600 8095 net.cpp:51] Initializing net from parameters:
...
...
...
layer {
name: "fc1x"
type: "InnerProduct"
bottom: "blstm_sum"
top: "fc1x"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 5990
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
axis: 2
}
}
layer {
name: "ctcloss"
type: "WarpCTCLoss"
bottom: "fc1x"
bottom: "label"
top: "ctcloss"
loss_weight: 1
}
I0515 15:14:44.814074 8095 layer_factory.hpp:77] Creating layer data
I0515 15:14:44.814433 8095 db_lmdb.cpp:35] Opened lmdb /home/zkgao/caffe_ocr_for_linux/data/crnn/lmdb/train_lmdb
I0515 15:14:44.814546 8095 net.cpp:84] Creating Layer data
I0515 15:14:44.814615 8095 net.cpp:380] data -> data
I0515 15:14:44.814857 8095 net.cpp:380] data -> label
*** Aborted at 1557904484 (unix time) try "date -d @1557904484" if you are using GNU date ***
PC: @ 0x7f155c152168 (unknown)
*** SIGSEGV (@0x15e7df10) received by PID 8095 (TID 0x7f15702667c0) from PID 367517456; stack trace: ***
@ 0x7f155c0394b0 (unknown)
@ 0x7f155c152168 (unknown)
@ 0x7f156e37ab07 std::__cxx11::basic_string<>::_M_construct<>()
@ 0x7f156e37ac05 std::__cxx11::basic_string<>::basic_string()
@ 0x7f156f5c6016 _ZN5caffe2db10LMDBCursor5valueB5cxx11Ev
@ 0x7f156f681fb0 caffe::DataLayer<>::DataLayerSetUp()
@ 0x7f156f752dbd caffe::BaseDataLayer<>::LayerSetUp()
@ 0x7f156f753213 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
@ 0x7f156f5f53ec caffe::Layer<>::SetUp()
@ 0x7f156f63859e caffe::Net<>::Init()
@ 0x7f156f636ff4 caffe::Net<>::Net()
@ 0x7f156f62aa03 caffe::Solver<>::InitTrainNet()
@ 0x7f156f62a225 caffe::Solver<>::Init()
@ 0x7f156f629d45 caffe::Solver<>::Solver()
@ 0x7f156f5e1322 caffe::SGDSolver<>::SGDSolver()
@ 0x7f156f5ed4b1 caffe::Creator_SGDSolver<>()
@ 0x424ac8 caffe::SolverRegistry<>::CreateSolver()
@ 0x41fc26 train()
@ 0x4221c5 main
@ 0x7f155c024830 __libc_start_main
@ 0x41e6c9 _start
@ 0x0 (unknown)
’
请问这种情况你遇到过吗

将fc1x层的节点数由5990变为36（因为我只识别数字和大写英文字母），无法训练，loss=nan

不定长标签无法收敛

训练标签长度包含2位3位4位5位6位7位时，网络收敛些不好，acc1的指标一直只有0.6，必须要定长么？