PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition

[giscus[bot]]

PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition

https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html

[immango]

使用“印章文本识别产线”无法替换模型

我在使用"印章文本识别产线"时，通过 paddlex --get_pipeline_config seal_recognition --save_path ./my_path 获取了配置文件，由于测试过程中文本识别模块准确率不达预期，所以我想通过替换文本识别模型为 PP-OCRv4_server_rec_doc来测试准确率是否会提升。修改后的配置文件：

Global:
pipeline_name: seal_recognition
input: https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png

Pipeline:
layout_model: RT-DETR-H_layout_3cls
text_det_model: PP-OCRv4_server_seal_det
text_rec_model: PP-OCRv4_server_rec_doc
layout_batch_size: 1
text_rec_batch_size: 1
此时重新运行程序时会出现报错：Exception: The model (PP-OCRv4_server_rec_doc) is no exists! Please using directory of local model files or model name supported by PaddleX!

但是我看PP-OCRv4_server_rec_doc这个模型应该是seal_recognition这条产线所支持的，是否哪里配置不对？

[qianqianer369]

请问您自己的文本识别模型，使用多大的数据集来训练的啊？效果怎么样啊

[qianqianer369]

官方提供的示例，好像只使用了版面检测模块RT-DETR-H_layout_3cls，印章文本检测模块PP-OCRv4_server_seal_det，文本识别模块PP-OCRv4_server_rec。 文本图像矫正模块，文档图像分类这些模块都没有使用。

[Punisher-pro]

请问怎么重新加载修改后的产线配置文件？

[Punisher-pro]

哦哦看到了解决了

[wuyongx]

请问下 通用OCR产线 高性能服务化部署 ，推荐的 服务器配置 是什么？最低配置是什么 ？

感谢

[cuicheng01]

这个没办法推荐，因为确实不知道最低的配置应该是什么，大多数的GPU都是可以的

[HuangShiKaiKai]

请问可以修改seal类别为其他类别吗

[sgliang1]

请问50系显卡预计什么时候可以适配呢

[yyyyyxie]

数据量很大，不适合下载到本地的话，如何改写Dataset类进行微调呢

[tianfeng912]

请问这个pipeline是不是线程安全的？：
from paddlex import create_pipeline
pipeline = create_pipeline(pipeline='OCR')

[lsq8377]

为什么我用python示例代码，识别不了示例图片的文字，返回的文本结果是空的.
下面是结果：
{
"input_path": "img/general_ocr_002.png",
"page_index": null,
"model_settings": {
"use_doc_preprocessor": false,
"use_textline_orientation": false
},
"dt_polys": [],
"text_det_params": {
"limit_side_len": 64,
"limit_type": "min",
"thresh": 0.3,
"max_side_limit": 4000,
"box_thresh": 0.6,
"unclip_ratio": 1.5
},
"text_type": "general",
"text_rec_score_thresh": 0.0,
"rec_texts": [],
"rec_scores": [],
"rec_polys": [],
"rec_boxes": []
}

[lsq8377]

from paddlex import create_pipeline

pipeline = create_pipeline(pipeline="OCR")

output = pipeline.predict(
input="img/general_ocr_002.png",
use_doc_orientation_classify=False,
use_doc_unwarping=False,
use_textline_orientation=False,
)
for res in output:
res.print()
res.save_to_img(save_path="./output/")
res.save_to_json(save_path="./output/")
这是代码

[zzzengzhe]

from paddlex import create_pipeline

pipeline = create_pipeline(pipeline="OCR")

在创建OCR管道时添加优化配置

output = pipeline.predict(
input="./page_1.png",
use_doc_orientation_classify=False,
use_doc_unwarping=False,
use_textline_orientation=False,
)
for res in output:
res.print()
res.save_to_img(save_path="./output/")
res.save_to_json(save_path="./output/")

Traceback (most recent call last):
File "/Contract_Review/src/utils/ocr.py", line 12, in
for res in output:
File "/root/PaddleX/paddlex/inference/pipelines/_parallel.py", line 129, in predict
yield from self._pipeline.predict(
File "/root/PaddleX/paddlex/inference/pipelines/ocr/pipeline.py", line 350, in predict
det_results = list(
File "/root/PaddleX/paddlex/inference/models/base/predictor/base_predictor.py", line 211, in call
yield from self.apply(input, **kwargs)
File "/root/PaddleX/paddlex/inference/models/base/predictor/base_predictor.py", line 267, in apply
prediction = self.process(batch_data, **kwargs)
File "/root/PaddleX/paddlex/inference/models/text_detection/predictor.py", line 105, in process
batch_preds = self.infer(x=x)
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 287, in call
pred = self.infer(x)
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 252, in call
self.predictor.run()
MemoryError: In user code:

ResourceExhaustedError: Fail to alloc memory of 14598144000 size, error code is 12.
  [Hint: Expected error == 0, but received error:12 != 0:0.] (at /paddle/paddle/phi/core/memory/allocation/cpu_allocator.cc:44)
  [operator < pd_kernel.phi_kernel > error]

使用cpu推理的话需要很大的内存吗？