在进行文本检测、识别串联推理时，出现错误 #250

hsyy673150343 · 2020-06-22T04:00:39Z

输入以下命令：
python3 tools/infer/predict_system.py --image_dir=/home/hs/OCR_Projects/PaddleOCR/train_data/tibetan/text_localization/test_image/img_9.jpg --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/"
出现以下错误：
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 118, in
dt_boxes, rec_res = text_sys(img)
File "tools/infer/predict_system.py", line 81, in call
rec_res, elapse = self.text_recognizer(img_crop_list)
File "/home/hs/OCR_Projects/PaddleOCR/tools/infer/predict_rec.py", line 101, in call
preds_text = self.char_ops.decode(rec_idx_tmp)
File "tools/infer/../../ppocr/utils/character.py", line 97, in decode
char_list.append(self.character[text_index[idx]])
IndexError: list index out of range

tink2123 · 2020-06-22T05:27:29Z

应该是因为预测模型使用的是中文模型，解码时指定了英文的字符解码方式。请问您使用的是哪一个识别模型呢？

hsyy673150343 · 2020-06-22T08:17:33Z

您好！我使用的识别模型是rcnn发自我的华为手机-------- 原始邮件 --------发件人： xiaoting <[email protected]>日期： 2020年6月22日周一下午1:27收件人： PaddlePaddle/PaddleOCR <[email protected]>抄送： "[email protected]" <[email protected]>, Author <[email protected]>主题： Re: [PaddlePaddle/PaddleOCR] 在进行文本检测、识别串联推理时，出现错误 (#250) 应该是因为预测模型使用的是中文模型，解码时指定了英文的字符解码方式。请问您使用的是哪一个识别模型呢？ —You are receiving this because you authored the thread.Reply to this email directly, view it on GitHub, or unsubscribe.

tink2123 · 2020-06-22T08:22:31Z

是中文模型吗？请试试预测时重新指定下解码方式：
python3 tools/infer/predict_system.py --image_dir=/home/hs/OCR_Projects/PaddleOCR/train_data/tibetan/text_localization/test_image/img_9.jpg --det_model_dir="./inference/det_db/" --rec_char_type='ch'

hsyy673150343 · 2020-06-22T08:41:21Z

是中文模型吗？请试试预测时重新指定下解码方式：
python3 tools/infer/predict_system.py --image_dir=/home/hs/OCR_Projects/PaddleOCR/train_data/tibetan/text_localization/test_image/img_9.jpg --det_model_dir="./inference/det_db/" --rec_char_type='ch'

python3 tools/infer/predict_system.py --image_dir=./doc/imgs/img_9.jpg --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --rec_char_type='ch'
还是会出现错误：
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 118, in
dt_boxes, rec_res = text_sys(img)
File "tools/infer/predict_system.py", line 81, in call
rec_res, elapse = self.text_recognizer(img_crop_list)
File "/home/hs/OCR_Projects/PaddleOCR/tools/infer/predict_rec.py", line 102, in call
preds_text = self.char_ops.decode(rec_idx_tmp)
File "tools/infer/../../ppocr/utils/character.py", line 96, in decode
char_list.append(self.character[text_index[idx]])
IndexError: list index out of range

tink2123 · 2020-06-22T08:55:21Z

打印下解码时的字典个数和预测的idx看一下呢？

PaddleOCR/ppocr/utils/character.py

Line 81 in f051be8

char_num = self.get_char_num()

在81行后：print(char_num)

PaddleOCR/ppocr/utils/character.py

Line 96 in f051be8

char_list.append(self.character[text_index[idx]])

96行前：print("text_index:{}, idx:{}".format(text_index,idx))

hsyy673150343 · 2020-06-22T09:08:03Z

print("text_index:{}, idx:{}".format(text_index,idx))

char_num: 6623
text_index:[82262 81698 82095 0 82262 82249 81770 81698 82262 81698 82262 81698
82009 81698 82262 82149 81698], idx:0

tink2123 · 2020-06-22T09:10:50Z

inference model 是从官方下载的吗？看起来预测结果有误，中文dict共有6623类，但是预测出了索引值都在8w+

hsyy673150343 · 2020-06-22T09:15:59Z

我是自己训练的模型然后按照文档里的说明转换成的inference model！
我的中文字典也是自定义的，我在rec_icdar15_train.yml中也指定了character_dict_path
但从char_num: 6623可以看出，程序还是用的自带的ppcor_keys_v1.txt

tink2123 · 2020-06-22T09:23:24Z

如果使用了自己的字典文件，在使用system预测时，需要修改字典路径，预测时添加： --rec_char_dict_path="{your text dict path}"

hsyy673150343 · 2020-06-22T09:41:51Z

谢谢您的耐心解答，我还有个问题。同一张图片，为什么我单独进行检测的时候能正常检测出文字，而在进行文本检测、识别串联的时候连最基本的检测都检测不出来？

tink2123 · 2020-06-22T09:44:26Z

您的图片是什么样子的呢，方便发一张样图给我们看下吗？

连最基本的检测都检测不出来

是指没有检测结果，还是检测结果错误呢

tink2123 · 2020-06-22T09:58:59Z

可能是因为识别得分低于0.5，置信度过低没有返回结果。可以把这个地方的阈值设置为0.0 看是否能显示出结果：

PaddleOCR/tools/infer/predict_system.py

Line 139 in f051be8

image, boxes, txts, scores, draw_txt=True, drop_score=0.5)

hsyy673150343 · 2020-06-22T10:02:54Z

可能是因为识别得分低于0.5，置信度过低没有返回结果。可以把这个地方的阈值设置为0.0 看是否能显示出结果：

PaddleOCR/tools/infer/predict_system.py

Line 139 in f051be8

image, boxes, txts, scores, draw_txt=True, drop_score=0.5)

改了之后报错啦：
Predict time of ./doc/imgs_words/ch/img_9.jpg: 1.049s
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 137, in
image, boxes, txts, scores, draw_txt=True, drop_score=0)
File "tools/infer/../../tools/infer/utility.py", line 154, in draw_ocr
txts, scores, img_h=img.shape[0], img_w=600, threshold=drop_score)
File "tools/infer/../../tools/infer/utility.py", line 241, in text_visual
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 351, in text
draw_text(ink)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 316, in draw_text
**kwargs,
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageFont.py", line 471, in getmask2
size, offset = self.font.getsize(text, direction, features, language)
OSError: invalid face handle
段错误 (核心已转储)

LDOUBLEV · 2020-06-22T11:19:45Z

可能是因为识别得分低于0.5，置信度过低没有返回结果。可以把这个地方的阈值设置为0.0 看是否能显示出结果：

PaddleOCR/tools/infer/predict_system.py

Line 139 in f051be8

image, boxes, txts, scores, draw_txt=True, drop_score=0.5)

改了之后报错啦：
Predict time of ./doc/imgs_words/ch/img_9.jpg: 1.049s
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 137, in
image, boxes, txts, scores, draw_txt=True, drop_score=0)
File "tools/infer/../../tools/infer/utility.py", line 154, in draw_ocr
txts, scores, img_h=img.shape[0], img_w=600, threshold=drop_score)
File "tools/infer/../../tools/infer/utility.py", line 241, in text_visual
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 351, in text
draw_text(ink)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 316, in draw_text
**kwargs,
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageFont.py", line 471, in getmask2
size, offset = self.font.getsize(text, direction, features, language)
OSError: invalid face handle
段错误 (核心已转储)

你是只测试识别模型？imgs_words/ch下的是识别数据，高度太小。测试检测用/doc/imgs下的图像试试？？

hsyy673150343 · 2020-06-22T12:37:49Z

可能是因为识别得分低于0.5，置信度过低没有返回结果。可以把这个地方的阈值设置为0.0 看是否能显示出结果：

PaddleOCR/tools/infer/predict_system.py

Line 139 in f051be8

image, boxes, txts, scores, draw_txt=True, drop_score=0.5)

改了之后报错啦：
Predict time of ./doc/imgs_words/ch/img_9.jpg: 1.049s
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 137, in
image, boxes, txts, scores, draw_txt=True, drop_score=0)
File "tools/infer/../../tools/infer/utility.py", line 154, in draw_ocr
txts, scores, img_h=img.shape[0], img_w=600, threshold=drop_score)
File "tools/infer/../../tools/infer/utility.py", line 241, in text_visual
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 351, in text
draw_text(ink)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 316, in draw_text
**kwargs,
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageFont.py", line 471, in getmask2
size, offset = self.font.getsize(text, direction, features, language)
OSError: invalid face handle
段错误 (核心已转储)

你是只测试识别模型？imgs_words/ch下的是识别数据，高度太小。测试检测用/doc/imgs下的图像试试？？

测试的是检测和识别串联推理，我用的测试命令如下：
python3 tools/infer/predict_system.py --image_dir=./doc/de_rec_test_img/img_9.jpg --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --rec_char_type="ch" --rec_char_dict_path="./ppocr/utils/all_tibetan_word.txt"

测试的图片是我自己收集的数据集，图片大小为608*608

报错信息为：
Traceback (most recent call last):
File "tools/infer/predict_system.py", line 139, in
image, boxes, txts, scores, draw_txt=True, drop_score=0.0)
File "tools/infer/../../tools/infer/utility.py", line 154, in draw_ocr
txts, scores, img_h=img.shape[0], img_w=600, threshold=drop_score)
File "tools/infer/../../tools/infer/utility.py", line 241, in text_visual
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 351, in text
draw_text(ink)
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageDraw.py", line 316, in draw_text
**kwargs,
File "/home/hs/.local/lib/python3.6/site-packages/PIL/ImageFont.py", line 471, in getmask2
size, offset = self.font.getsize(text, direction, features, language)
OSError: invalid face handle
段错误 (核心已转储)

tink2123 · 2020-06-22T12:42:41Z

报错说明返回了检测框，在画图了。这好像是Pillow库的bug，可能因为您的字典中有特殊字符？可以参考这个issue尝试重新安装编译：python-pillow/Pillow#3066

或者跳过draw_text函数: draw_txt=False

PaddleOCR/tools/infer/utility.py

Line 311 in f051be8

new_img = draw_ocr(image, boxes, txts, scores, draw_txt=True)

返回结果可以通过修改阈值，在终端打印出来，score>=0.0 ：

PaddleOCR/tools/infer/predict_system.py

Line 127 in f051be8

if score >= 0.5:

hsyy673150343 · 2020-06-22T15:00:16Z

报错说明返回了检测框，在画图了。这好像是Pillow库的bug，可能因为您的字典中有特殊字符？可以参考这个issue尝试重新安装编译：python-pillow/Pillow#3066

或者跳过draw_text函数: draw_txt=False

PaddleOCR/tools/infer/utility.py

Line 311 in f051be8

new_img = draw_ocr(image, boxes, txts, scores, draw_txt=True)

返回结果可以通过修改阈值，在终端打印出来，score>=0.0 ：

PaddleOCR/tools/infer/predict_system.py

Line 127 in f051be8

if score >= 0.5:

谢谢您的耐心解答~！我的问题解决了。
我首先卸载了pillow----》pip uninstall pillow==7.1.2
然后将score改为0
最后使用conda重新下载了pillow----》conda install pillow==7.1.2
这个bug就没有了

tink2123 closed this as completed Jun 23, 2020

jitesh-rathod mentioned this issue Dec 14, 2020

段错误 (核心已转储) #1238

Closed

an1018 pushed a commit to an1018/PaddleOCR that referenced this issue Aug 17, 2022

Fix load checkpoint (PaddlePaddle#250)

3849c17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

在进行文本检测、识别串联推理时，出现错误 #250

在进行文本检测、识别串联推理时，出现错误 #250

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020 via email

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

tink2123 commented Jun 22, 2020 •

edited

Loading

hsyy673150343 commented Jun 22, 2020

LDOUBLEV commented Jun 22, 2020 •

edited

Loading

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

在进行文本检测、识别串联推理时，出现错误 #250

在进行文本检测、识别串联推理时，出现错误 #250

Comments

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020 via email

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

tink2123 commented Jun 22, 2020 • edited Loading

hsyy673150343 commented Jun 22, 2020

LDOUBLEV commented Jun 22, 2020 • edited Loading

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020

hsyy673150343 commented Jun 22, 2020

tink2123 commented Jun 22, 2020 •

edited

Loading

LDOUBLEV commented Jun 22, 2020 •

edited

Loading