CTPN是什么

CTPN結(jié)合CNN與LSTM深度網(wǎng)絡(luò)勇垛，CTPN是從Faster R-CNN改進(jìn)而來(lái)味榛，能有效的檢測(cè)出復(fù)雜場(chǎng)景的橫向分布的文字，效果如圖1媚媒，是目前比較好的文字檢測(cè)算法篓像。詳細(xì)解釋?zhuān)?a target="_blank" rel="nofollow">傳送門(mén)

image.png

本次要實(shí)現(xiàn)的項(xiàng)目

地址如下（別著急克屡钠狻）：

https://github.com/eragonruan/text-detection-ctpn

image.png

項(xiàng)目要實(shí)現(xiàn)的效果

image.png

image.png

第一步--克隆項(xiàng)目

我克隆的是下面這個(gè)項(xiàng)目，他有訓(xùn)練好的權(quán)重文件：

https://github.com/GuoYi0/text_detection

第二步--開(kāi)始跑demo

結(jié)果碰到了下面這個(gè)錯(cuò)誤：

ImportError: cannot import name 'bbox'

回去找issue屈暗，果然有人跟我一樣拆讯，才看半截就開(kāi)跑
issue地址：https://github.com/eragonruan/text-detection-ctpn/issues/59
下面有人給出了答案：

image.png

照著做脂男，再跑一次，但是种呐，宰翅，，還是

image.png

another error：

Invalid argument: ValueError: Buffer dtype mismatch, expected 'int_t' but got 'long long'

再去找解決方案爽室。找到了下面這個(gè)：
issue地址：https://github.com/eragonruan/text-detection-ctpn/issues/59
解決方案如下（我只是部分修改汁讼，原作者是僅使用CPU執(zhí)行）

thanks to the author and [#43](https://github.com/eragonruan/text-detection-ctpn/issues/43) zhao181

my environment is:
windows10 ,
python3.6 ,
tensorflow1.3 ,
vs2015(ps:vs2013 not support python3.6 when compile)

step 1:make some change
change "np.int_t " to "np.intp_t" in line 25 of the file lib\utils\cython_nms.pyx
otherwise appear " ValueError: Buffer dtype mismatch, expected 'int_t' but got 'long long' " in step 6.

step 2:updata c file
execute:cd your_dir\text-detection-ctpn-master\lib\utils
execute:cython bbox.pyx
execute:cython cython_nms.pyx

step 3:builf setup file as setup_new.py
import numpy as np
from distutils.core import setup
from Cython.Build import cythonize
from distutils.extension import Extension
numpy_include = np.get_include()
setup(ext_modules=cythonize("bbox.pyx"),include_dirs=[numpy_include])
setup(ext_modules=cythonize("cython_nms.pyx"),include_dirs=[numpy_include])

step 4:build .pyd file
execute:python setup_new.py install
copy bbox.cp36-win_amd64.pyd and cython_nms.cp36-win_amd64.pyd to your_dir\text-detection-ctpn-master\lib\utils

step 5:make some change
(1) Set "USE_GPU_NMS " in the file \ctpn\text.yml as "False"
(2) Set the "_*C.USE_GPU_NMS" in the file \lib\fast_rcnn\config.py as "False";
(3) Comment out the line "from lib.utils.gpu_nms import gpu_nms" in the file \lib\fast_rcnn\nms_wrapper.py;
(4) Comment out the line "from . import gpu_nms" in the file \lib\utils_*init**.py;
(5) change "base_name = image_name.split('/')[-1]" to "base_name = image_name.split('\')[-1]" in line 24 of the file ctpn\demo.py

step 6:run demo
execute:cd your_dir\text-detection-ctpn-master
execute:python ./ctpn/demo.py

關(guān)鍵改動(dòng)部分

step 1:make some change
change "np.int_t " to "np.intp_t" in line 25 of the file lib\utils\cython_nms.pyx
otherwise appear " ValueError: Buffer dtype mismatch, expected 'int_t' but got 'long long' " in step 6.

我再跑一次測(cè)試！
結(jié)果阔墩，嘿架，，

image.png

錯(cuò)誤信息如下：

Loading network VGGnet_test...  Restoring from checkpoints/VGGnet_fast_rcnn_iter_50000.ckpt... done
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for E:\business\recognition\text_detection\text-detection-ctpn\data\demo\010.png
Traceback (most recent call last):
  File "./ctpn/demo.py", line 101, in <module>
    ctpn(sess, net, im_name)
  File "./ctpn/demo.py", line 61, in ctpn
    draw_boxes(img, image_name, boxes, scale)
  File "./ctpn/demo.py", line 26, in draw_boxes
    with open('data/results/' + 'res_{}.txt'.format(base_name.split('.')[0]), 'w') as f:
OSError: [Errno 22] Invalid argument: 'data/results/res_E:\\business\\recognition\\text_detection\\text-detection-ctpn\\data\\demo\\010.txt'

定位這個(gè)問(wèn)題花了我一點(diǎn)時(shí)間啸箫，各種猜想耸彪，最后再回頭詳細(xì)讀一下這個(gè)錯(cuò)誤信息，發(fā)現(xiàn)了問(wèn)題所在忘苛，最后一句蝉娜！參數(shù)錯(cuò)誤！
回去改源碼：

 base_name = image_name.split('\\')[-1]
    with open('data\\results\\' + 'res_{}.txt'.format(base_name.split('.')[0]), 'w') as f:

只想說(shuō)一句扎唾，Windows的“\”真是召川。。胸遇。

image.png

image.png

image.png

image.png

再補(bǔ)一句届案，對(duì)這個(gè)感興趣的還可以去研究下下面這個(gè)項(xiàng)目庵楷，提供了數(shù)據(jù)集等

https://github.com/QAlexBall/Faster_RCNN_for_TextDetection