Merge branch 'dygraph' into dygraph

PPOCRLabel/libs/shape.py

@@ -48,6 +48,7 @@ class Shape(object):
 
     def __init__(self, label=None, line_color=None, difficult=False, key_cls="None", paintLabel=False):
         self.label = label
+        self.idx = 0
         self.points = []
         self.fill = False
         self.selected = False

README_ch.md

@@ -71,6 +71,8 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力
 ## 《动手学OCR》电子书
 - [《动手学OCR》电子书📚](./doc/doc_ch/ocr_book.md)
 
+## 场景应用
+- PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。详情可查看[README](./applications)。
 
 <a name="开源社区"></a>
 ## 开源社区

applications/PCB字符识别/PCB字符识别.md

@@ -206,7 +206,11 @@ Eval.dataset.transforms.DetResizeForTest:  尺寸
         limit_type: 'min'
 ```
 
-然后执行评估代码
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型评估。
 
 
 ```python

applications/README.md

+# 场景应用
+
+PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
+
+> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
+
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
+
+## 通用
+
+| 类别                                              | 亮点     | 类别       | 亮点         |
+| ------------------------------------------------- | -------- | ---------- | ------------ |
+| [高精度中文识别模型SVTR](./高精度中文识别模型.md) | 新增模型 | 手写体识别 | 新增字形支持 |
+
+## 制造
+
+| 类别                                                         | 亮点                           | 类别                                        | 亮点                 |
+| ------------------------------------------------------------ | ------------------------------ | ------------------------------------------- | -------------------- |
+| [数码管识别](./光功率计数码管字符识别/光功率计数码管字符识别.md) | 数码管数据合成、漏识别调优     | 电表识别                                    | 大分辨率图像检测调优 |
+| [液晶屏读数识别](./液晶屏读数识别.md)                        | 检测模型蒸馏、Serving部署      | [PCB文字识别](./PCB字符识别/PCB字符识别.md) | 小尺寸文本检测与识别 |
+| [包装生产日期](./包装生产日期识别.md)                        | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测                              | 非文字字符识别       |
+
+## 金融
+
+| 类别                           | 亮点                     | 类别         | 亮点                  |
+| ------------------------------ | ------------------------ | ------------ | --------------------- |
+| [表单VQA](./多模态表单识别.md) | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票                     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别                 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+
+## 交通
+
+| 类别                            | 亮点                           | 类别       | 亮点     |
+| ------------------------------- | ------------------------------ | ---------- | -------- |
+| [车牌识别](./轻量级车牌识别.md) | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别               | 尽请期待                       |            |          |
\ No newline at end of file

applications/光功率计数码管字符识别/corpus/digital.txt

+46.39
+40.08
+89.52
+-71.93
+23.19
+-81.02
+-34.09
+05.87
+-67.80
+-51.56
+-34.58
+37.91
+56.98
+29.01
+-90.13
+35.55
+66.07
+-90.35
+-50.93
+42.42
+21.40
+-30.99
+-71.78
+25.60
+-48.69
+-72.28
+-17.55
+-99.93
+-47.35
+-64.89
+-31.28
+-90.01
+05.17
+30.91
+30.56
+-06.90
+79.05
+67.74
+-32.31
+94.22
+28.75
+51.03
+-58.96

applications/光功率计数码管字符识别/光功率计数码管字符识别.md

+# 光功率计数码管字符识别
+
+本案例将使用OCR技术自动识别光功率计显示屏文字，通过本章您可以掌握：
+
+- PaddleOCR快速使用
+- 数据合成方法
+- 数据挖掘方法
+- 基于现有数据微调
+
+## 1. 背景介绍
+
+光功率计（optical power meter ）是指用于测量绝对光功率或通过一段光纤的光功率相对损耗的仪器。在光纤系统中，测量光功率是最基本的，非常像电子学中的万用表；在光纤测量中，光功率计是重负荷常用表。
+
+<img src="https://bkimg.cdn.bcebos.com/pic/a08b87d6277f9e2f999f5e3e1c30e924b899f35a?x-bce-process=image/watermark,image_d2F0ZXIvYmFpa2U5Mg==,g_7,xp_5,yp_5/format,f_auto" width="400">
+
+目前光功率计缺少将数据直接输出的功能，需要人工读数。这一项工作单调重复，如果可以使用机器替代人工，将节约大量成本。针对上述问题，希望通过摄像头拍照->智能读数的方式高效地完成此任务。
+
+为实现智能读数，通常会采取文本检测+文本识别的方案：
+
+第一步，使用文本检测模型定位出光功率计中的数字部分；
+
+第二步，使用文本识别模型获得准确的数字和单位信息。
+
+本项目主要介绍如何完成第二步文本识别部分，包括：真实评估集的建立、训练数据的合成、基于 PP-OCRv3 和 SVTR_Tiny 两个模型进行训练，以及评估和推理。
+
+本项目难点如下：
+
+- 光功率计数码管字符数据较少，难以获取。
+- 数码管中小数点占像素较少，容易漏识别。
+
+针对以上问题， 本例选用 PP-OCRv3 和 SVTR_Tiny 两个高精度模型训练，同时提供了真实数据挖掘案例和数据合成案例。基于 PP-OCRv3 模型，在构建的真实评估集上精度从 52% 提升至 72%，SVTR_Tiny 模型精度可达到 78.9%。
+
+aistudio项目链接: [光功率计数码管字符识别](https://aistudio.baidu.com/aistudio/projectdetail/4049044?contributionType=1)
+
+## 2. PaddleOCR 快速使用
+
+PaddleOCR 旨在打造一套丰富、领先、且实用的OCR工具库，助力开发者训练出更好的模型，并应用落地。
+
+![](https://github.com/PaddlePaddle/PaddleOCR/raw/release/2.5/doc/imgs_results/ch_ppocr_mobile_v2.0/test_add_91.jpg)
+
+
+官方提供了适用于通用场景的高精轻量模型，首先使用官方提供的 PP-OCRv3 模型预测图片，验证下当前模型在光功率计场景上的效果。
+
+- 准备环境
+
+```
+python3 -m pip install -U pip
+python3 -m pip install paddleocr
+```
+
+
+- 测试效果
+
+测试图：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/8dca91f016884e16ad9216d416da72ea08190f97d87b4be883f15079b7ebab9a)
+
+
+```
+paddleocr --lang=ch --det=Fase --image_dir=data
+```
+
+得到如下测试结果：
+
+```
+('.7000', 0.6885431408882141)
+```
+
+发现数字识别较准，然而对负号和小数点识别不准确。 由于PP-OCRv3的训练数据大多为通用场景数据，在特定的场景上效果可能不够好。因此需要基于场景数据进行微调。
+
+下面就主要介绍如何在光功率计（数码管）场景上微调训练。
+
+
+## 3. 开始训练
+
+### 3.1 数据准备
+
+特定的工业场景往往很难获取开源的真实数据集，光功率计也是如此。在实际工业场景中，可以通过摄像头采集的方法收集大量真实数据，本例中重点介绍数据合成方法和真实数据挖掘方法，如何利用有限的数据优化模型精度。
+
+数据集分为两个部分：合成数据，真实数据， 其中合成数据由 text_renderer 工具批量生成得到， 真实数据通过爬虫等方式在百度图片中搜索并使用 PPOCRLabel 标注得到。
+
+
+- 合成数据
+
+本例中数据合成工具使用的是 [text_renderer](https://github.com/Sanster/text_renderer)， 该工具可以合成用于文本识别训练的文本行数据:
+
+![](https://github.com/oh-my-ocr/text_renderer/raw/master/example_data/effect_layout_image/char_spacing_compact.jpg)
+
+![](https://github.com/oh-my-ocr/text_renderer/raw/master/example_data/effect_layout_image/color_image.jpg)
+
+
+```
+export https_proxy=http://172.19.57.45:3128
+git clone https://github.com/oh-my-ocr/text_renderer
+```
+
+```
+import os
+python3 setup.py develop
+python3 -m pip install -r docker/requirements.txt
+python3 main.py \
+    --config example_data/example.py \
+    --dataset img \
+    --num_processes 2 \
+    --log_period 10
+```
+
+给定字体和语料，就可以合成较为丰富样式的文本行数据。 光功率计识别场景，目标是正确识别数码管文本，因此需要收集部分数码管字体，训练语料，用于合成文本识别数据。
+
+将收集好的语料存放在 example_data 路径下：
+
+```
+ln -s ./fonts/DS* text_renderer/example_data/font/
+ln -s ./corpus/digital.txt text_renderer/example_data/text/
+```
+
+修改 text_renderer/example_data/font_list/font_list.txt ,选择需要的字体开始合成：
+
+```
+python3 main.py \
+    --config example_data/digital_example.py \
+    --dataset img \
+    --num_processes 2 \
+    --log_period 10
+```
+
+合成图片会被存在目录 text_renderer/example_data/digital/chn_data 下
+
+查看合成的数据样例：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/7d5774a273f84efba5b9ce7fd3f86e9ef24b6473e046444db69fa3ca20ac0986)
+
+
+- 真实数据挖掘
+
+模型训练需要使用真实数据作为评价指标，否则很容易过拟合到简单的合成数据中。没有开源数据的情况下，可以利用部分无标注数据+标注工具获得真实数据。
+
+
+1. 数据搜集
+
+使用[爬虫工具](https://github.com/Joeclinton1/google-images-download.git)获得无标注数据
+
+2. [PPOCRLabel](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5/PPOCRLabel) 完成半自动标注
+
+PPOCRLabel是一款适用于OCR领域的半自动化图形标注工具，内置PP-OCR模型对数据自动标注和重新识别。使用Python3和PyQT5编写，支持矩形框标注、表格标注、不规则文本标注、关键信息标注模式，导出格式可直接用于PaddleOCR检测和识别模型的训练。
+
+![](https://github.com/PaddlePaddle/PaddleOCR/raw/release/2.5/PPOCRLabel/data/gif/steps_en.gif)
+
+
+收集完数据后就可以进行分配了，验证集中一般都是真实数据，训练集中包含合成数据+真实数据。本例中标注了155张图片，其中训练集和验证集的数目为100和55。
+
+
+最终 `data` 文件夹应包含以下几部分：
+
+```
+|-data
+  |- synth_train.txt
+  |- real_train.txt
+  |- real_eval.txt
+  |- synthetic_data
+      |- word_001.png
+      |- word_002.jpg
+      |- word_003.jpg
+      | ...
+  |- real_data
+      |- word_001.png
+      |- word_002.jpg
+      |- word_003.jpg
+      | ...
+  ...
+```
+
+### 3.2 模型选择
+
+本案例提供了2种文本识别模型：PP-OCRv3 识别模型 和 SVTR_Tiny：
+
+[PP-OCRv3 识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md)：PP-OCRv3的识别模块是基于文本识别算法SVTR优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。并进行了一系列结构改进加速模型预测。
+
+[SVTR_Tiny](https://arxiv.org/abs/2205.00159):SVTR提出了一种用于场景文本识别的单视觉模型，该模型在patch-wise image tokenization框架内，完全摒弃了序列建模，在精度具有竞争力的前提下，模型参数量更少，速度更快。
+
+以上两个策略在自建中文数据集上的精度和速度对比如下：
+
+| ID | 策略 |  模型大小 | 精度 | 预测耗时（CPU + MKLDNN)|
+|-----|-----|--------|----| --- |
+| 01 | PP-OCRv2 | 8M | 74.8% | 8.54ms |
+| 02 | SVTR_Tiny | 21M | 80.1% | 97ms |
+| 03 | SVTR_LCNet(h32) | 12M | 71.9% | 6.6ms |
+| 04 | SVTR_LCNet(h48) | 12M | 73.98% | 7.6ms |
+| 05 | + GTC | 12M | 75.8% | 7.6ms |
+| 06 | + TextConAug | 12M | 76.3% | 7.6ms |
+| 07 | + TextRotNet | 12M | 76.9% | 7.6ms |
+| 08 | + UDML | 12M | 78.4% | 7.6ms |
+| 09 | + UIM | 12M | 79.4% | 7.6ms |
+
+
+### 3.3 开始训练
+
+首先下载 PaddleOCR 代码库
+
+```
+git clone -b release/2.5 https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 PP-OCRv3 中文识别模型为例：
+
+**Step1：下载预训练模型**
+
+首先下载 pretrain model，您可以下载训练好的模型在自定义数据上进行finetune
+
+```
+cd PaddleOCR/
+# 下载PP-OCRv3 中文预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+# 解压模型参数
+cd pretrain_models
+tar -xf ch_PP-OCRv3_rec_train.tar && rm -rf ch_PP-OCRv3_rec_train.tar
+```
+
+**Step2：自定义字典文件**
+
+接下来需要提供一个字典（{word_dict_name}.txt），使模型在训练时，可以将所有出现的字符映射为字典的索引。
+
+因此字典需要包含所有希望被正确识别的字符，{word_dict_name}.txt需要写成如下格式，并以 `utf-8` 编码格式保存：
+
+```
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+-
+.
+```
+
+word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，“3.14” 将被映射成 [3, 11, 1, 4]
+
+* 内置字典
+
+PaddleOCR内置了一部分字典，可以按需使用。
+
+`ppocr/utils/ppocr_keys_v1.txt` 是一个包含6623个字符的中文字典
+
+`ppocr/utils/ic15_dict.txt` 是一个包含36个字符的英文字典
+
+* 自定义字典
+
+内置字典面向通用场景，具体的工业场景中，可能需要识别特殊字符，或者只需识别某几个字符，此时自定义字典会更提升模型精度。例如在光功率计场景中，需要识别数字和单位。
+
+遍历真实数据标签中的字符，制作字典`digital_dict.txt`如下所示：
+
+```
+-
+.
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+B
+E
+F
+H
+L
+N
+T
+W
+d
+k
+m
+n
+o
+z
+```
+
+
+
+
+**Step3：修改配置文件**
+
+为了更好的使用预训练模型，训练推荐使用[ch_PP-OCRv3_rec_distillation.yml](../../configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml)配置文件，并参考下列说明修改配置文件：
+
+以 `ch_PP-OCRv3_rec_distillation.yml` 为例：
+```
+Global:
+  ...
+  # 添加自定义字典，如修改字典请将路径指向新字典
+  character_dict_path: ppocr/utils/dict/digital_dict.txt
+  ...
+  # 识别空格
+  use_space_char: True
+
+
+Optimizer:
+  ...
+  # 添加学习率衰减策略
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  ...
+
+...
+
+Train:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./data/
+    # 训练集标签文件
+    label_file_list:
+    - ./train_data/digital_img/digital_train.txt  #11w
+    - ./train_data/digital_img/real_train.txt     #100
+    - ./train_data/digital_img/dbm_img/dbm.txt    #3w
+    ratio_list:
+    - 0.3
+    - 1.0
+    - 1.0
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    ...
+    # 单卡训练的batch_size
+    batch_size_per_card: 256
+    ...
+
+Eval:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./data
+    # 验证集标签文件
+    label_file_list:
+    - ./train_data/digital_img/real_val.txt
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    # 单卡验证的batch_size
+    batch_size_per_card: 256
+    ...
+```
+**注意，训练/预测/评估时的配置文件请务必与训练一致。**
+
+**Step4：启动训练**
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```
+# GPU训练 支持单卡，多卡训练
+# 训练数码管数据 训练日志会自动保存为 "{save_model_dir}" 下的train.log
+
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model=./pretrain_models/ch_PP-OCRv3_rec_train/best_accuracy
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy
+```
+
+
+PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml` 中修改 `eval_batch_step` 设置评估频率，默认每500个iter评估一次。评估过程中默认将最佳acc模型，保存为 `output/ch_PP-OCRv3_rec_distill/best_accuracy` 。
+
+如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
+
+### SVTR_Tiny 训练
+
+SVTR_Tiny 训练步骤与上面一致，SVTR支持的配置和模型训练权重可以参考[算法介绍文档](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/algorithm_rec_svtr.md)
+
+**Step1：下载预训练模型**
+
+```
+# 下载 SVTR_Tiny 中文识别预训练模型和配置文件
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_ch_train.tar
+# 解压模型参数
+tar -xf rec_svtr_tiny_none_ctc_ch_train.tar && rm -rf rec_svtr_tiny_none_ctc_ch_train.tar
+```
+**Step2：自定义字典文件**
+
+字典依然使用自定义的 digital_dict.txt
+
+**Step3：修改配置文件**
+
+配置文件中对应修改字典路径和数据路径
+
+**Step4：启动训练**
+
+```
+## 单卡训练
+python tools/train.py -c rec_svtr_tiny_none_ctc_ch_train/rec_svtr_tiny_6local_6global_stn_ch.yml \
+           -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_ch_train/best_accuracy
+```
+
+### 3.4 验证效果
+
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理
+
+* 指标评估
+
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。评估数据集可以通过 `configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml`  修改Eval中的 `label_file_path` 设置。
+
+```
+# GPU 评估， Global.checkpoints 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+* 测试识别效果
+
+使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。
+
+默认预测图片存储在 `infer_img` 里，通过 `-o Global.checkpoints` 加载训练好的参数文件：
+
+根据配置文件中设置的 `save_model_dir` 和 `save_epoch_step` 字段，会有以下几种参数被保存下来：
+
+```
+output/rec/
+├── best_accuracy.pdopt  
+├── best_accuracy.pdparams  
+├── best_accuracy.states  
+├── config.yml  
+├── iter_epoch_3.pdopt  
+├── iter_epoch_3.pdparams  
+├── iter_epoch_3.states  
+├── latest.pdopt  
+├── latest.pdparams  
+├── latest.states  
+└── train.log
+```
+
+其中 best_accuracy.* 是评估集上的最优模型；iter_epoch_x.* 是以 `save_epoch_step` 为间隔保存下来的模型；latest.* 是最后一个epoch的模型。
+
+```
+# 预测英文结果
+python3 tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.infer_img=test_digital.png
+```
+
+预测图片：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/8dca91f016884e16ad9216d416da72ea08190f97d87b4be883f15079b7ebab9a)
+
+
+得到输入图像的预测结果：
+
+```
+infer_img: test_digital.png
+        result: ('-70.00', 0.9998967)
+```

applications/轻量级车牌识别.md

@@ -311,7 +311,6 @@ python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o
 
 在上述命令中，通过`-o`的方式修改了配置文件中的参数。
 
-训练好的模型地址为: [det_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar)
 
 **评估**
 
@@ -354,8 +353,6 @@ python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCR
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
 ```
 
-训练好的模型地址为: [det_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar)
-
 量化后指标对比如下
 
 |方案|hmeans| 模型大小 | 预测速度(lite) |
@@ -436,6 +433,12 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
 
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
 评估部分日志如下：
 ```bash
 [2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
@@ -528,7 +531,6 @@ python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
-训练好的模型地址为: [rec_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar)
 
 **评估**
 
@@ -570,7 +572,6 @@ python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
-训练好的模型地址为: [rec_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar)
 
 量化后指标对比如下
 

deploy/avh/.gitignore

+include/inputs.h
+include/outputs.h
+
+__pycache__/
+build/
\ No newline at end of file

deploy/avh/Makefile

+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+# Makefile to build demo
+
+# Setup build environment
+BUILD_DIR := build
+
+ARM_CPU = ARMCM55
+ETHOSU_PATH = /opt/arm/ethosu
+CMSIS_PATH ?= ${ETHOSU_PATH}/cmsis
+ETHOSU_PLATFORM_PATH ?= ${ETHOSU_PATH}/core_platform
+STANDALONE_CRT_PATH := $(abspath $(BUILD_DIR))/runtime
+CORSTONE_300_PATH = ${ETHOSU_PLATFORM_PATH}/targets/corstone-300
+PKG_COMPILE_OPTS = -g -Wall -O2 -Wno-incompatible-pointer-types -Wno-format -mcpu=cortex-m55 -mthumb -mfloat-abi=hard -std=gnu99
+CMAKE ?= cmake
+CC = arm-none-eabi-gcc
+AR = arm-none-eabi-ar
+RANLIB = arm-none-eabi-ranlib
+PKG_CFLAGS = ${PKG_COMPILE_OPTS} \
+	-I${STANDALONE_CRT_PATH}/include \
+	-I${STANDALONE_CRT_PATH}/src/runtime/crt/include \
+	-I${PWD}/include \
+	-I${CORSTONE_300_PATH} \
+	-I${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Include/ \
+	-I${CMSIS_PATH}/CMSIS/Core/Include \
+	-I${CMSIS_PATH}/CMSIS/NN/Include \
+	-I${CMSIS_PATH}/CMSIS/DSP/Include \
+	-I$(abspath $(BUILD_DIR))/codegen/host/include
+CMSIS_NN_CMAKE_FLAGS = -DCMAKE_TOOLCHAIN_FILE=$(abspath $(BUILD_DIR))/../arm-none-eabi-gcc.cmake \
+	-DTARGET_CPU=cortex-m55 \
+	-DBUILD_CMSIS_NN_FUNCTIONS=YES
+PKG_LDFLAGS = -lm -specs=nosys.specs -static -T corstone300.ld
+
+$(ifeq VERBOSE,1)
+QUIET ?=
+$(else)
+QUIET ?= @
+$(endif)
+
+DEMO_MAIN = src/demo_bare_metal.c
+CODEGEN_SRCS = $(wildcard $(abspath $(BUILD_DIR))/codegen/host/src/*.c)
+CODEGEN_OBJS = $(subst .c,.o,$(CODEGEN_SRCS))
+CMSIS_STARTUP_SRCS = $(wildcard ${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Source/*.c)
+UART_SRCS = $(wildcard ${CORSTONE_300_PATH}/*.c)
+
+demo: $(BUILD_DIR)/demo
+
+$(BUILD_DIR)/stack_allocator.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/memory/stack_allocator.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+$(BUILD_DIR)/crt_backend_api.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/common/crt_backend_api.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+# Build generated code
+$(BUILD_DIR)/libcodegen.a: $(CODEGEN_SRCS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/codegen/host/src) && $(CC) -c $(PKG_CFLAGS) $(CODEGEN_SRCS)
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcodegen.a) $(CODEGEN_OBJS)
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcodegen.a)
+
+# Build CMSIS startup code
+${BUILD_DIR}/libcmsis_startup.a: $(CMSIS_STARTUP_SRCS)
+	$(QUIET)mkdir -p $(abspath $(BUILD_DIR)/libcmsis_startup)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/libcmsis_startup) && $(CC) -c $(PKG_CFLAGS) -D${ARM_CPU} $^
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcmsis_startup.a) $(abspath $(BUILD_DIR))/libcmsis_startup/*.o
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcmsis_startup.a)
+
+CMSIS_SHA_FILE=${CMSIS_PATH}/977abe9849781a2e788b02282986480ff4e25ea6.sha
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+else
+# Build CMSIS-NN
+${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+endif
+
+# Build demo application
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+	${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a ${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+else
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+       ${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a \
+       ${BUILD_DIR}/cmsis_nn/Source/FullyConnectedFunctions/libCMSISNNFullyConnected.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SVDFunctions/libCMSISNNSVDF.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ReshapeFunctions/libCMSISNNReshape.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ActivationFunctions/libCMSISNNActivation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/NNSupportFunctions/libCMSISNNSupport.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConcatenationFunctions/libCMSISNNConcatenation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/BasicMathFunctions/libCMSISNNBasicMaths.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConvolutionFunctions/libCMSISNNConvolutions.a \
+       ${BUILD_DIR}/cmsis_nn/Source/PoolingFunctions/libCMSISNNPooling.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+endif
+
+clean:
+	$(QUIET)rm -rf $(BUILD_DIR)/codegen
+
+cleanall:
+	$(QUIET)rm -rf $(BUILD_DIR)
+
+.SUFFIXES:
+
+.DEFAULT: demo

deploy/avh/README.md

+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+Running PaddleOCR text recognition model via TVM on bare metal Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
+===============================================================
+
+This folder contains an example of how to use TVM to run a PaddleOCR model
+on bare metal Cortex(R)-M55 CPU and CMSIS-NN.
+
+Prerequisites
+-------------
+If the demo is run in the ci_cpu Docker container provided with TVM, then the following
+software will already be installed.
+
+If the demo is not run in the ci_cpu Docker container, then you will need the following:
+- Software required to build and run the demo (These can all be installed by running 
+  https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosu_driver_stack.sh .)
+  - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+  - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+  - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+  - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+  - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- The python libraries listed in the requirements.txt of this directory
+  - These can be installed by running the following from the current directory:
+    ```bash
+    pip install -r ./requirements.txt
+    ```
+
+You will also need TVM which can either be:
+  - Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    - When building from source, the following need to be set in config.cmake:
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+  - Installed from TLCPack nightly(see [TLCPack](https://tlcpack.ai/))
+
+You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.
+For example if you've installed these in ```/opt/arm``` , then you would do the following:
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+Running the demo application
+----------------------------
+Type the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):
+
+```bash
+./run_demo.sh
+```
+
+If the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then
+the locations for these can be specified as arguments to run_demo.sh, for example:
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+This will:
+- Download a PaddleOCR text recognition model
+- Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN
+- Create a C header file inputs.c containing the image data as a C array
+- Create a C header file outputs.c containing a C array where the output of inference will be stored
+- Build the demo application
+- Run the demo application on a Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software
+- The application will report the text on the image and the corresponding score.
+
+Using your own image
+--------------------
+The create_image.py script takes a single argument on the command line which is the path of the
+image to be converted into an array of bytes for consumption by the model.
+
+The demo can be modified to use an image of your choice by changing the following line in run_demo.sh
+
+```bash
+python3 ./convert_image.py path/to/image
+```
+
+Model description
+-----------------
+In this demo, the model we use is an English recognition model based on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md). PP-OCRv3 is the third version of the PP-OCR series model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). This series of models has the following features:
+  - PP-OCRv3: ultra-lightweight OCR system: detection (3.6M) + direction classifier (1.4M) + recognition (12M) = 17.0M
+  - Support more than 80 kinds of multi-language recognition models, including English, Chinese, French, German, Arabic, Korean, Japanese and so on. For details
+  - Support vertical text recognition, and long text recognition
+
+The text recognition model in PP-OCRv3 supports more than 80 languages. In the process of model development, since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current model.
\ No newline at end of file

deploy/avh/README_ch.md

+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+通过TVM在 Arm(R) Cortex(R)-M55 CPU 上运行 PaddleOCR文 本能识别模型
+===============================================================
+
+此文件夹包含如何使用 TVM 在 Cortex(R)-M55 CPU 上运行 PaddleOCR 模型的示例。
+
+依赖
+-------------
+本demo运行在TVM提供的docker环境上，在该环境中已经安装好的必须的软件
+
+
+在非docker环境中，需要手动安装如下依赖项: 
+
+- 软件可通过[安装脚本](https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosu_driver_stack.sh)一键安装
+  - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+  - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+  - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+  - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+  - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- python 依赖
+  ```bash
+  pip install -r ./requirements.txt
+  ```
+- TVM
+  - 从源码安装([Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    从源码安装时，需要设置如下字段
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+  - 从TLCPack 安装([TLCPack](https://tlcpack.ai/))
+
+安装完成后需要更新环境变量，以软件安装地址为`/opt/arm`为例：
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+运行demo
+----------------------------
+使用如下命令可以一键运行demo
+
+```bash
+./run_demo.sh
+```
+
+如果 Ethos(TM)-U 平台或 CMSIS 没有安装在 `/opt/arm/ethosu` 中，可通过参数进行设置，例如：
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+`./run_demo.sh`脚本会执行如下步骤:
+- 下载 PaddleOCR 文字识别模型
+- 使用tvm将PaddleOCR 文字识别模型编译为 Cortex(R)-M55 CPU 和 CMSIS-NN 后端的可执行文件
+- 创建一个包含输入图像数据的头文件`inputs.c`
+- 创建一个包含输出tensor大小的头文件`outputs.c`
+- 编译可执行程序
+- 运行程序
+- 输出图片上的文字和置信度
+
+使用自己的图片
+--------------------
+替换 `run_demo.sh ` 中140行处的图片地址即可
+
+使用自己的模型
+--------------------
+替换 `run_demo.sh ` 中130行处的模型地址即可
+
+模型描述
+-----------------
+
+在这个demo中，我们使用的模型是基于[PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md)的英文识别模型。 PP-OCRv3是[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)发布的PP-OCR系列模型的第三个版本。 该系列模型具有以下特点：
+   - 超轻量级OCR系统：检测（3.6M）+方向分类器（1.4M）+识别（12M）=17.0M。
+   - 支持80多种多语言识别模型，包括英文、中文、法文、德文、阿拉伯文、韩文、日文等。 
+   - 支持竖排文本识别，长文本识别。
+
+PP-OCRv3 中的文本识别模型支持 80 多种语言。 在模型开发过程中，由于Arm(R) Cortex(R)-M55 CPU不支持rnn算子，我们在PP-OCRv3文本识别模型的基础上删除了不支持的算子，得到当前模型。
\ No newline at end of file

deploy/avh/arm-none-eabi-gcc.cmake

+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+if (__TOOLCHAIN_LOADED)
+    return()
+endif()
+set(__TOOLCHAIN_LOADED TRUE)
+
+set(CMAKE_SYSTEM_NAME Generic)
+set(CMAKE_C_COMPILER "arm-none-eabi-gcc")
+set(CMAKE_CXX_COMPILER "arm-none-eabi-g++")
+set(CMAKE_SYSTEM_PROCESSOR "cortex-m55" CACHE STRING "Select Arm(R) Cortex(R)-M architecture. (cortex-m0, cortex-m3, cortex-m33, cortex-m4, cortex-m55, cortex-m7, etc)")
+
+set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+
+SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_CXX_STANDARD 14)
+
+# The system processor could for example be set to cortex-m33+nodsp+nofp.
+set(__CPU_COMPILE_TARGET ${CMAKE_SYSTEM_PROCESSOR})
+string(REPLACE "+" ";" __CPU_FEATURES ${__CPU_COMPILE_TARGET})
+list(POP_FRONT __CPU_FEATURES CMAKE_SYSTEM_PROCESSOR)
+
+string(FIND ${__CPU_COMPILE_TARGET} "+" __OFFSET)
+if(__OFFSET GREATER_EQUAL 0)
+    string(SUBSTRING ${__CPU_COMPILE_TARGET} ${__OFFSET} -1 CPU_FEATURES)
+endif()
+
+# Add -mcpu to the compile options to override the -mcpu the CMake toolchain adds
+add_compile_options(-mcpu=${__CPU_COMPILE_TARGET})
+
+# Set floating point unit
+if("${__CPU_COMPILE_TARGET}" MATCHES "\\+fp")
+    set(FLOAT hard)
+elseif("${__CPU_COMPILE_TARGET}" MATCHES "\\+nofp")
+    set(FLOAT soft)
+elseif("${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m33" OR
+       "${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m55")
+    set(FLOAT hard)
+else()
+    set(FLOAT soft)
+endif()
+
+add_compile_options(-mfloat-abi=${FLOAT})
+add_link_options(-mfloat-abi=${FLOAT})
+
+# Link target
+add_link_options(-mcpu=${__CPU_COMPILE_TARGET})
+add_link_options(-Xlinker -Map=output.map)
+
+#
+# Compile options
+#
+set(cxx_flags "-fno-unwind-tables;-fno-rtti;-fno-exceptions")
+
+add_compile_options("-Wall;-Wextra;-Wsign-compare;-Wunused;-Wswitch-default;\
+-Wdouble-promotion;-Wredundant-decls;-Wshadow;-Wnull-dereference;\
+-Wno-format-extra-args;-Wno-unused-function;-Wno-unused-label;\
+-Wno-missing-field-initializers;-Wno-return-type;-Wno-format;-Wno-int-conversion"
+    "$<$<COMPILE_LANGUAGE:CXX>:${cxx_flags}>"
+)

deploy/avh/convert_image.py

+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+import os
+import pathlib
+import re
+import sys
+import cv2
+import math
+from PIL import Image
+import numpy as np
+
+def resize_norm_img(img, image_shape, padding=True):
+    imgC, imgH, imgW = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    if not padding:
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_w = imgW
+    else:
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype('float32')
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+    padding_im[:, :, 0:resized_w] = resized_image
+    return padding_im
+
+
+def create_header_file(name, tensor_name, tensor_data, output_path):
+    """
+    This function generates a header file containing the data from the numpy array provided.
+    """
+    file_path = pathlib.Path(f"{output_path}/" + name).resolve()
+    # Create header file with npy_data as a C array
+    raw_path = file_path.with_suffix(".h").resolve()
+    with open(raw_path, "w") as header_file:
+        header_file.write(
+            "\n"
+            + f"const size_t {tensor_name}_len = {tensor_data.size};\n"
+            + f'__attribute__((section(".data.tvm"), aligned(16))) float {tensor_name}[] = '
+        )
+
+        header_file.write("{")
+        for i in np.ndindex(tensor_data.shape):
+            header_file.write(f"{tensor_data[i]}, ")
+        header_file.write("};\n\n")
+
+
+def create_headers(image_name):
+    """
+    This function generates C header files for the input and output arrays required to run inferences
+    """
+    img_path = os.path.join("./", f"{image_name}")
+
+    # Resize image to 32x320
+    img = cv2.imread(img_path)
+    img = resize_norm_img(img, [3,32,320])
+    img_data = img.astype("float32")
+    
+    # # Add the batch dimension, as we are expecting 4-dimensional input: NCHW.
+    img_data = np.expand_dims(img_data, axis=0)
+
+    # Create input header file
+    create_header_file("inputs", "input", img_data, "./include")
+    # Create output header file
+    output_data = np.zeros([7760], np.float)
+    create_header_file(
+        "outputs",
+        "output",
+        output_data,
+        "./include",
+    )
+
+
+if __name__ == "__main__":
+    create_headers(sys.argv[1])

deploy/avh/corstone300.ld

+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*------------------ Reference System Memories -------------
+  +===================+============+=======+============+============+
+  | Memory            | Address    | Size  | CPU Access | NPU Access |
+  +===================+============+=======+============+============+
+  | ITCM              | 0x00000000 | 512KB | Yes (RO)   | No         |
+  +-------------------+------------+-------+------------+------------+
+  | DTCM              | 0x20000000 | 512KB | Yes (R/W)  | No         |
+  +-------------------+------------+-------+------------+------------+
+  | SSE-300 SRAM      | 0x21000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | Data SRAM         | 0x01000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | DDR               | 0x60000000 |  32MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+ */
+
+/*---------------------- ITCM Configuration ----------------------------------
+  <h> Flash Configuration
+    <o0> Flash Base Address <0x0-0xFFFFFFFF:8>
+    <o1> Flash Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__ROM_BASE = 0x00000000;
+__ROM_SIZE = 0x00080000;
+
+/*--------------------- DTCM RAM Configuration ----------------------------
+  <h> RAM Configuration
+    <o0> RAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> RAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__RAM_BASE = 0x20000000;
+__RAM_SIZE = 0x00080000;
+
+/*----------------------- Data SRAM Configuration ------------------------------
+  <h> Data SRAM Configuration
+    <o0> DATA_SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DATA_SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DATA_SRAM_BASE = 0x01000000;
+__DATA_SRAM_SIZE = 0x00200000;
+
+/*--------------------- Embedded SRAM Configuration ----------------------------
+  <h> SRAM Configuration
+    <o0> SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__SRAM_BASE = 0x21000000;
+__SRAM_SIZE = 0x00200000;
+
+/*--------------------- Stack / Heap Configuration ----------------------------
+  <h> Stack / Heap Configuration
+    <o0> Stack Size (in Bytes) <0x0-0xFFFFFFFF:8>
+    <o1> Heap Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__STACK_SIZE = 0x00008000;
+__HEAP_SIZE  = 0x00008000;
+
+/*--------------------- Embedded RAM Configuration ----------------------------
+  <h> DDR Configuration
+    <o0> DDR Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DDR Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DDR_BASE = 0x60000000;
+__DDR_SIZE = 0x02000000;
+
+/*
+ *-------------------- <<< end of configuration section >>> -------------------
+ */
+
+MEMORY
+{
+  ITCM       (rx)  : ORIGIN = __ROM_BASE, LENGTH = __ROM_SIZE
+  DTCM       (rwx) : ORIGIN = __RAM_BASE, LENGTH = __RAM_SIZE
+  DATA_SRAM  (rwx) : ORIGIN = __DATA_SRAM_BASE, LENGTH = __DATA_SRAM_SIZE
+  SRAM       (rwx) : ORIGIN = __SRAM_BASE, LENGTH = __SRAM_SIZE
+  DDR        (rwx) : ORIGIN = __DDR_BASE, LENGTH = __DDR_SIZE
+}
+
+/* Linker script to place sections and symbol values. Should be used together
+ * with other linker script that defines memory regions ITCM and RAM.
+ * It references following symbols, which must be defined in code:
+ *   Reset_Handler : Entry of reset handler
+ *
+ * It defines following symbols, which code can use without definition:
+ *   __exidx_start
+ *   __exidx_end
+ *   __copy_table_start__
+ *   __copy_table_end__
+ *   __zero_table_start__
+ *   __zero_table_end__
+ *   __etext
+ *   __data_start__
+ *   __preinit_array_start
+ *   __preinit_array_end
+ *   __init_array_start
+ *   __init_array_end
+ *   __fini_array_start
+ *   __fini_array_end
+ *   __data_end__
+ *   __bss_start__
+ *   __bss_end__
+ *   __end__
+ *   end
+ *   __HeapLimit
+ *   __StackLimit
+ *   __StackTop
+ *   __stack
+ */
+ENTRY(Reset_Handler)
+
+SECTIONS
+{
+  /* .ddr is placed before .text so that .rodata.tvm is encountered before .rodata* */
+  .ddr :
+  {
+    . = ALIGN (16);
+    *(.rodata.tvm)
+    . = ALIGN (16);
+    *(.data.tvm);
+    . = ALIGN(16);    
+  } > DDR
+
+  .text :
+  {
+    KEEP(*(.vectors))
+    *(.text*)
+
+    KEEP(*(.init))
+    KEEP(*(.fini))
+
+    /* .ctors */
+    *crtbegin.o(.ctors)
+    *crtbegin?.o(.ctors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .ctors)
+    *(SORT(.ctors.*))
+    *(.ctors)
+
+    /* .dtors */
+    *crtbegin.o(.dtors)
+    *crtbegin?.o(.dtors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .dtors)
+    *(SORT(.dtors.*))
+    *(.dtors)
+
+    *(.rodata*)
+
+    KEEP(*(.eh_frame*))
+  } > ITCM
+
+  .ARM.extab :
+  {
+    *(.ARM.extab* .gnu.linkonce.armextab.*)
+  } > ITCM
+
+  __exidx_start = .;
+  .ARM.exidx :
+  {
+    *(.ARM.exidx* .gnu.linkonce.armexidx.*)
+  } > ITCM
+  __exidx_end = .;
+
+  .copy.table :
+  {
+    . = ALIGN(4);
+    __copy_table_start__ = .;
+    LONG (__etext)
+    LONG (__data_start__)
+    LONG (__data_end__ - __data_start__)
+    /* Add each additional data section here */
+    __copy_table_end__ = .;
+  } > ITCM
+
+  .zero.table :
+  {
+    . = ALIGN(4);
+    __zero_table_start__ = .;
+    __zero_table_end__ = .;
+  } > ITCM
+
+  /**
+   * Location counter can end up 2byte aligned with narrow Thumb code but
+   * __etext is assumed by startup code to be the LMA of a section in DTCM
+   * which must be 4byte aligned
+   */
+  __etext = ALIGN (4);
+
+  .sram :
+  {
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .data : AT (__etext)
+  {
+    __data_start__ = .;
+    *(vtable)
+    *(.data)
+    *(.data.*)
+
+    . = ALIGN(4);
+    /* preinit data */
+    PROVIDE_HIDDEN (__preinit_array_start = .);
+    KEEP(*(.preinit_array))
+    PROVIDE_HIDDEN (__preinit_array_end = .);
+
+    . = ALIGN(4);
+    /* init data */
+    PROVIDE_HIDDEN (__init_array_start = .);
+    KEEP(*(SORT(.init_array.*)))
+    KEEP(*(.init_array))
+    PROVIDE_HIDDEN (__init_array_end = .);
+
+
+    . = ALIGN(4);
+    /* finit data */
+    PROVIDE_HIDDEN (__fini_array_start = .);
+    KEEP(*(SORT(.fini_array.*)))
+    KEEP(*(.fini_array))
+    PROVIDE_HIDDEN (__fini_array_end = .);
+
+    KEEP(*(.jcr*))
+    . = ALIGN(4);
+    /* All data end */
+    __data_end__ = .;
+
+  } > DTCM
+
+  .bss.noinit (NOLOAD):
+  {
+    . = ALIGN(16);
+    *(.bss.noinit.*)
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .bss :
+  {
+    . = ALIGN(4);
+    __bss_start__ = .;
+    *(.bss)
+    *(.bss.*)
+    *(COMMON)
+    . = ALIGN(4);
+    __bss_end__ = .;
+  } > DTCM AT > DTCM
+
+  .data_sram :
+  {
+    . = ALIGN(16);
+  } > DATA_SRAM
+
+  .heap (COPY) :
+  {
+    . = ALIGN(8);
+    __end__ = .;
+    PROVIDE(end = .);
+    . = . + __HEAP_SIZE;
+    . = ALIGN(8);
+    __HeapLimit = .;
+  } > DTCM
+
+  .stack (ORIGIN(DTCM) + LENGTH(DTCM) - __STACK_SIZE) (COPY) :
+  {
+    . = ALIGN(8);
+    __StackLimit = .;
+    . = . + __STACK_SIZE;
+    . = ALIGN(8);
+    __StackTop = .;
+  } > DTCM
+  PROVIDE(__stack = __StackTop);
+
+  /* Check if data + stack exceeds DTCM limit */
+  ASSERT(__StackLimit >= __bss_end__, "region DTCM overflowed with stack")
+}

deploy/avh/include/crt_config.h

+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#ifndef TVM_RUNTIME_CRT_CONFIG_H_
+#define TVM_RUNTIME_CRT_CONFIG_H_
+
+/*! Log level of the CRT runtime */
+#define TVM_CRT_LOG_LEVEL TVM_CRT_LOG_LEVEL_DEBUG
+
+#endif  // TVM_RUNTIME_CRT_CONFIG_H_

deploy/avh/include/tvm_runtime.h

+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <tvm/runtime/c_runtime_api.h>
+#include <tvm/runtime/crt/stack_allocator.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void __attribute__((noreturn)) TVMPlatformAbort(tvm_crt_error_t error_code) {
+  printf("TVMPlatformAbort: %d\n", error_code);
+  printf("EXITTHESIM\n");
+  exit(-1);
+}
+
+tvm_crt_error_t TVMPlatformMemoryAllocate(size_t num_bytes, DLDevice dev, void** out_ptr) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+tvm_crt_error_t TVMPlatformMemoryFree(void* ptr, DLDevice dev) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+void TVMLogf(const char* msg, ...) {
+  va_list args;
+  va_start(args, msg);
+  vfprintf(stdout, msg, args);
+  va_end(args);
+}
+
+TVM_DLL int TVMFuncRegisterGlobal(const char* name, TVMFunctionHandle f, int override) { return 0; }
+
+#ifdef __cplusplus
+}
+#endif

deploy/avh/requirements.txt

+paddlepaddle
+numpy
+opencv-python
\ No newline at end of file

deploy/avh/run_demo.sh

+#!/bin/bash
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+set -e
+set -u
+set -o pipefail
+
+# Show usage
+function show_usage() {
+    cat <<EOF
+Usage: run_demo.sh
+-h, --help
+    Display this help message.
+--cmsis_path CMSIS_PATH
+    Set path to CMSIS.
+--ethosu_platform_path ETHOSU_PLATFORM_PATH
+    Set path to Arm(R) Ethos(TM)-U core platform.
+--fvp_path FVP_PATH
+   Set path to FVP.
+--cmake_path
+   Set path to cmake.
+EOF
+}
+
+# Parse arguments
+while (( $# )); do
+    case "$1" in
+        -h|--help)
+            show_usage
+            exit 0
+            ;;
+
+        --cmsis_path)
+            if [ $# -gt 1 ]
+            then
+                export CMSIS_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmsis_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --ethosu_platform_path)
+            if [ $# -gt 1 ]
+            then
+                export ETHOSU_PLATFORM_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --ethosu_platform_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --fvp_path)
+            if [ $# -gt 1 ]
+            then
+                export PATH="$2/models/Linux64_GCC-6.4:$PATH"
+                shift 2
+            else
+                echo 'ERROR: --fvp_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --cmake_path)
+            if [ $# -gt 1 ]
+            then
+                export CMAKE="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmake_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        -*|--*)
+            echo "Error: Unknown flag: $1" >&2
+            show_usage >&2
+            exit 1
+            ;;
+    esac
+done
+
+
+# Directories
+script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+
+
+# Make build directory
+rm -rf build
+make cleanall
+mkdir -p build
+cd build
+
+wget https://paddleocr.bj.bcebos.com/tvm/ocr_en.tar
+tar -xf ocr_en.tar
+
+# Compile model for Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
+# An alternative to using "python3 -m tvm.driver.tvmc" is to call
+# "tvmc" directly once TVM has been pip installed.
+python3 -m tvm.driver.tvmc compile --target=cmsis-nn,c \
+    --target-cmsis-nn-mcpu=cortex-m55 \
+    --target-c-mcpu=cortex-m55 \
+    --runtime=crt \
+    --executor=aot \
+    --executor-aot-interface-api=c \
+    --executor-aot-unpacked-api=1 \
+    --pass-config tir.usmp.enable=1 \
+    --pass-config tir.usmp.algorithm=hill_climb \
+    --pass-config tir.disable_storage_rewrite=1 \
+    --pass-config tir.disable_vectorize=1 ocr_en/inference.pdmodel \
+    --output-format=mlf \
+    --model-format=paddle \
+    --module-name=rec \
+    --input-shapes x:[1,3,32,320] \
+    --output=rec.tar
+tar -xf rec.tar
+
+# Create C header files
+cd ..
+python3 ./convert_image.py imgs_words_en/word_116.png
+
+# Build demo executable
+cd ${script_dir}
+echo ${script_dir}
+make
+
+# Run demo executable on the FVP
+FVP_Corstone_SSE-300_Ethos-U55 -C cpu0.CFGDTCMSZ=15 \
+-C cpu0.CFGITCMSZ=15 -C mps3_board.uart0.out_file=\"-\" -C mps3_board.uart0.shutdown_tag=\"EXITTHESIM\" \
+-C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 \
+-C mps3_board.telnetterminal1.start_telnet=0 -C mps3_board.telnetterminal2.start_telnet=0 -C mps3_board.telnetterminal5.start_telnet=0 \
+./build/demo
\ No newline at end of file

deploy/avh/src/demo_bare_metal.c

+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdio.h>
+#include <tvm_runtime.h>
+#include <tvmgen_rec.h>
+
+#include "uart.h"
+
+// Header files generated by convert_image.py
+#include "inputs.h"
+#include "outputs.h"
+
+
+int main(int argc, char** argv) {
+  char dict[]={"#0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~!\"#$%&'()*+,-./  "};
+  int char_dict_nums = 97;
+  uart_init();
+  printf("Starting ocr rec inference\n");
+  struct tvmgen_rec_outputs rec_outputs = {
+      .output = output,
+  };
+  struct tvmgen_rec_inputs rec_inputs = {
+      .x = input,
+  };
+
+  tvmgen_rec_run(&rec_inputs, &rec_outputs);
+
+  // post process
+  int char_nums = output_len / char_dict_nums;
+  
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+  
+  printf("text: ");
+  for (int i = 0; i < char_nums; i++) {
+    int argmax_idx = 0;
+    float max_value = 0.0f;
+    for (int j = 0; j < char_dict_nums; j++){
+      if (output[i * char_dict_nums + j] > max_value){
+        max_value = output[i * char_dict_nums + j];
+        argmax_idx = j;
+      }
+    }
+    if (argmax_idx > 0 && (!(i > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      // printf("%d，%f，%c\n", argmax_idx, max_value, dict[argmax_idx]);
+      printf("%c", dict[argmax_idx]);
+    }
+    last_index = argmax_idx;
+  }
+  score /= count;
+  printf(", score: %f\n", score);
+  
+  // The FVP will shut down when it receives "EXITTHESIM" on the UART
+  printf("EXITTHESIM\n");
+  while (1 == 1)
+    ;
+  return 0;
+}

deploy/lite/config.txt

@@ -4,4 +4,5 @@ det_db_box_thresh  0.5
 det_db_unclip_ratio  1.6
 det_db_use_dilate 0
 det_use_polygon_score 1
-use_direction_classify  1
\ No newline at end of file
+use_direction_classify  1
+rec_image_height  32
\ No newline at end of file

deploy/lite/crnn_process.cc

@@ -19,25 +19,27 @@
 
 const std::vector<int> rec_image_shape{3, 32, 320};
 
-cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio) {
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height) {
   int imgC, imgH, imgW;
   imgC = rec_image_shape[0];
+  imgH = rec_image_height;
   imgW = rec_image_shape[2];
-  imgH = rec_image_shape[1];
 
-  imgW = int(32 * wh_ratio);
+  imgW = int(imgH * wh_ratio);
 
-  float ratio = static_cast<float>(img.cols) / static_cast<float>(img.rows);
+  float ratio = float(img.cols) / float(img.rows);
   int resize_w, resize_h;
+
   if (ceilf(imgH * ratio) > imgW)
     resize_w = imgW;
   else
-    resize_w = static_cast<int>(ceilf(imgH * ratio));
-  cv::Mat resize_img;
+    resize_w = int(ceilf(imgH * ratio));
+
   cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
              cv::INTER_LINEAR);
-
-  return resize_img;
+  cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
 }
 
 std::vector<std::string> ReadDict(std::string path) {

deploy/lite/crnn_process.h

@@ -26,7 +26,7 @@
 #include "opencv2/imgcodecs.hpp"
 #include "opencv2/imgproc.hpp"
 
-cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio);
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height);
 
 std::vector<std::string> ReadDict(std::string path);
 

deploy/lite/ocr_db_crnn.cc

@@ -162,7 +162,8 @@ void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
                  std::vector<std::string> charactor_dict,
                  std::shared_ptr<PaddlePredictor> predictor_cls,
                  int use_direction_classify,
-                 std::vector<double> *times) {
+                 std::vector<double> *times,
+                 int rec_image_height) {
   std::vector<float> mean = {0.5f, 0.5f, 0.5f};
   std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
 
@@ -183,7 +184,7 @@ void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
     float wh_ratio =
         static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
 
-    resize_img = CrnnResizeImg(crop_img, wh_ratio);
+    resize_img = CrnnResizeImg(crop_img, wh_ratio, rec_image_height);
     resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
 
     const float *dimg = reinterpret_cast<const float *>(resize_img.data);
@@ -444,7 +445,7 @@ void system(char **argv){
   //// load config from txt file
   auto Config = LoadConfigTxt(det_config_path);
   int use_direction_classify = int(Config["use_direction_classify"]);
-
+  int rec_image_height = int(Config["rec_image_height"]);
   auto charactor_dict = ReadDict(dict_path);
   charactor_dict.insert(charactor_dict.begin(), "#"); // blank char for ctc
   charactor_dict.push_back(" ");
@@ -590,12 +591,16 @@ void rec(int argc, char **argv) {
   std::string batchsize = argv[6];
   std::string img_dir = argv[7];
   std::string dict_path = argv[8];
+  std::string config_path = argv[9];
 
   if (strcmp(argv[4], "FP32") != 0 && strcmp(argv[4], "INT8") != 0) {
       std::cerr << "Only support FP32 or INT8." << std::endl;
       exit(1);
   }
 
+  auto Config = LoadConfigTxt(config_path);
+  int rec_image_height = int(Config["rec_image_height"]);
+
   std::vector<cv::String> cv_all_img_names;
   cv::glob(img_dir, cv_all_img_names);
 
@@ -630,7 +635,7 @@ void rec(int argc, char **argv) {
     std::vector<float> rec_text_score;
     std::vector<double> times;
     RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
-                charactor_dict, cls_predictor, 0, &times);
+                charactor_dict, cls_predictor, 0, &times, rec_image_height);
   
     //// print recognized text
     for (int i = 0; i < rec_text.size(); i++) {

deploy/lite/readme.md

@@ -34,7 +34,7 @@ For the compilation process of different development environments, please refer
 ### 1.2 Prepare Paddle-Lite library
 
 There are two ways to obtain the Paddle-Lite library：
-- 1. Download directly, the download link of the Paddle-Lite library is as follows：
+- 1. [Recommended] Download directly, the download link of the Paddle-Lite library is as follows：
 
       | Platform | Paddle-Lite library download link |
       |---|---|
@@ -43,7 +43,9 @@ There are two ways to obtain the Paddle-Lite library：
 
       Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
 
-- 2. [Recommended] Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
+    **Note: It is recommended to use paddlelite>=2.10 version of the prediction library, other prediction library versions [download link](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -104,21 +106,17 @@ If you directly use the model in the above table for deployment, you can skip th
 
 If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
 
-The `opt` tool can be obtained by compiling Paddle Lite.
+- Step 1: Refer to [document](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html) to install paddlelite, which is used to convert paddle inference model to paddlelite required for running nb model
 ```
-git clone https://github.com/PaddlePaddle/Paddle-Lite.git
-cd Paddle-Lite
-git checkout release/v2.10
-./lite/tools/build.sh build_optimize_tool
+pip install paddlelite==2.10 # The paddlelite version should be the same as the prediction library version
 ```
-
-After the compilation is complete, the opt file is located under build.opt/lite/api/, You can view the operating options and usage of opt in the following ways:
-
+After installation, the following commands can view the help information
 ```
-cd build.opt/lite/api/
-./opt
+paddle_lite_opt
 ```
 
+Introduction to paddle_lite_opt parameters:
+
 |Options|Description|
 |---|---|
 |--model_dir|The path of the PaddlePaddle model to be optimized (non-combined form)|
@@ -131,6 +129,8 @@ cd build.opt/lite/api/
 
 `--model_dir` is suitable for the non-combined mode of the model to be optimized, and the inference model of PaddleOCR is the combined mode, that is, the model structure and model parameters are stored in a single file.
 
+- Step 2: Use paddle_lite_opt to convert the inference model to the mobile model format.
+
 The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
 
 ```
@@ -240,6 +240,7 @@ det_db_thresh  0.3        # Used to filter the binarized image of DB prediction,
 det_db_box_thresh  0.5    # DDB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate
 det_db_unclip_ratio  1.6  # Indicates the compactness of the text box, the smaller the value, the closer the text box to the text
 use_direction_classify  0  # Whether to use the direction classifier, 0 means not to use, 1 means to use
+rec_image_height  32      # The height of the input image of the recognition model, the PP-OCRv3 model needs to be set to 48, and the PP-OCRv2 model needs to be set to 32
 ```
 
  5. Run Model on phone
@@ -258,8 +259,15 @@ After the above steps are completed, you can use adb to push the file to the pho
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # The use of ocr_db_crnn is:
- # ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file Test image path Dictionary file path
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ # ./ocr_db_crnn Mode Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
+ ./ocr_db_crnn system ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+# precision can be INT8 for quantitative model or FP32 for normal model.
+
+# Only using detection model
+./ocr_db_crnn  det ch_PP-OCRv2_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# Only using recognition model
+./ocr_db_crnn  rec ch_PP-OCRv2_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
  ```
 
 If you modify the code, you need to recompile and push to the phone.
@@ -283,3 +291,7 @@ A2: Replace the .jpg test image under ./debug with the image you want to test, a
 Q3: How to package it into the mobile APP?
 
 A3: This demo aims to provide the core algorithm part that can run OCR on mobile phones. Further, PaddleOCR/deploy/android_demo is an example of encapsulating this demo into a mobile app for reference.
+
+Q4: When running the demo, an error is reported `Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4: The problem is that the installed paddlelite version does not match the downloaded prediction library version. Make sure that the paddleliteopt tool matches your prediction library version, and try to switch to the nb model again.

deploy/lite/readme_ch.md

@@ -8,7 +8,7 @@
     - [2.1 模型优化](#21-模型优化)
     - [2.2 与手机联调](#22-与手机联调)
 - [FAQ](#faq)
-  
+
 
 本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
 
@@ -32,7 +32,7 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 ### 1.2 准备预测库
 
 预测库有两种获取方式：
-- 1. 直接下载，预测库下载链接如下：
+- 1. [推荐]直接下载，预测库下载链接如下：
 
       | 平台 | 预测库下载链接 |
       |---|---|
@@ -41,7 +41,9 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 
       注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
 
-- 2. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
+**注：建议使用paddlelite>=2.10版本的预测库，其他预测库版本[下载链接](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. 编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -102,22 +104,16 @@ Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括
 
 如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
 
-模型优化需要Paddle-Lite的opt可执行文件，可以通过编译Paddle-Lite源码获得，编译步骤如下：
+- 步骤1：参考[文档](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html)安装paddlelite，用于转换paddle inference model为paddlelite运行所需的nb模型
 ```
-# 如果准备环境时已经clone了Paddle-Lite，则不用重新clone Paddle-Lite
-git clone https://github.com/PaddlePaddle/Paddle-Lite.git
-cd Paddle-Lite
-git checkout release/v2.10
-# 启动编译
-./lite/tools/build.sh build_optimize_tool
+pip install paddlelite==2.10  # paddlelite版本要与预测库版本一致
 ```
-
-编译完成后，opt文件位于`build.opt/lite/api/`下，可通过如下方式查看opt的运行选项和使用方式；
+安装完后，如下指令可以查看帮助信息
 ```
-cd build.opt/lite/api/
-./opt
+paddle_lite_opt
 ```
 
+paddle_lite_opt 参数介绍：
 |选项|说明|
 |---|---|
 |--model_dir|待优化的PaddlePaddle模型（非combined形式）的路径|
@@ -130,6 +126,8 @@ cd build.opt/lite/api/
 
 `--model_dir`适用于待优化的模型是非combined方式，PaddleOCR的inference模型是combined方式，即模型结构和模型参数使用单独一个文件存储。
 
+- 步骤2：使用paddle_lite_opt将inference模型转换成移动端模型格式。
+
 下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
 
 ```
@@ -148,7 +146,7 @@ wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls
 
 转换成功后，inference模型目录下会多出`.nb`结尾的文件，即是转换成功的模型文件。
 
-注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt 工具的输入模型是paddle保存的inference模型
+注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt工具的输入模型是paddle保存的inference模型
 
 <a name="2.2与手机联调"></a>
 ### 2.2 与手机联调
@@ -234,13 +232,14 @@ ppocr_keys_v1.txt   # 中文字典
 ...
 ```
 
-2.  `config.txt` 包含了检测器、分类器的超参数，如下：
+2.  `config.txt` 包含了检测器、分类器、识别器的超参数，如下：
 ```
 max_side_len  960         # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960
 det_db_thresh  0.3        # 用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显
-det_db_box_thresh  0.5    # DB后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
+det_db_box_thresh  0.5    # 检测器后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
 det_db_unclip_ratio  1.6  # 表示文本框的紧致程度，越小则文本框更靠近文本
 use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1表示使用
+rec_image_height  32      # 识别模型输入图像的高度，PP-OCRv3模型设置为48，PP-OCRv2模型需要设置为32
 ```
 
  5. 启动调试
@@ -259,8 +258,14 @@ use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # 开始使用，ocr_db_crnn可执行文件的使用方式为:
- # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件  测试图像路径  字典文件路径
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ # ./ocr_db_crnn 预测模式  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
+ ./ocr_db_crnn system  ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+
+# 仅使用文本检测模型，使用方式如下：
+./ocr_db_crnn  det ch_PP-OCRv2_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# 仅使用文本识别模型，使用方式如下：
+./ocr_db_crnn  rec ch_PP-OCRv2_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
  ```
 
  如果对代码做了修改，则需要重新编译并push到手机上。
@@ -284,3 +289,7 @@ A2：替换debug下的.jpg测试图像为你想要测试的图像，adb push 到
 Q3：如何封装到手机APP中？
 
 A3：此demo旨在提供能在手机上运行OCR的核心算法部分，PaddleOCR/deploy/android_demo是将这个demo封装到手机app的示例，供参考
+
+Q4：运行demo时遇到报错`Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4：问题是安装的paddlelite版本和下载的预测库版本不匹配，确保paddleliteopt工具和你的预测库版本匹配，重新转nb模型试试。

doc/doc_ch/PPOCRv3_det_train.md

+
+# PP-OCRv3 文本检测模型训练
+
+- [1. 简介](#1)
+- [2. PPOCRv3检测训练](#2)
+- [3. 基于PPOCRv3检测的finetune训练](#3)
+
+<a name="1"></a>
+## 1. 简介
+
+PP-OCRv3在PP-OCRv2的基础上进一步升级。本节介绍PP-OCRv3检测模型的训练步骤。有关PPOCRv3策略介绍参考[文档](./PP-OCRv3_introduction.md)。
+
+
+<a name="2"></a>
+## 2. 检测训练
+
+PP-OCRv3检测模型是对PP-OCRv2中的[CML](https://arxiv.org/pdf/2109.03144.pdf)（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。PP-OCRv3分别针对检测教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+
+PP-OCRv3检测训练包括两个步骤：
+- 步骤1：采用DML蒸馏方法训练检测教师模型
+- 步骤2：使用步骤1得到的教师模型采用CML方法训练出轻量学生模型
+
+
+### 2.1 准备数据和运行环境
+
+训练数据采用icdar2015数据，准备训练集步骤参考[ocr_dataset](./dataset/ocr_datasets.md).
+
+运行环境准备参考[文档](./installation.md)。
+
+
+### 2.2 训练教师模型
+
+教师模型训练的配置文件是[ch_PP-OCRv3_det_dml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml)。教师模型模型结构的Backbone、Neck、Head分别为Resnet50, LKPAN, DBHead，采用DML的蒸馏方法训练。有关配置文件的详细介绍参考[文档](./knowledge_distillation)。
+
+
+下载ImageNet预训练模型：
+```
+# 下载ResNet50_vd的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+```
+
+**启动训练**
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，包含以下文件：
+```
+best_accuracy.states    
+best_accuracy.pdparams  # 默认保存最优精度的模型参数
+best_accuracy.pdopt     # 默认保存最优精度的优化器相关参数
+latest.states    
+latest.pdparams  # 默认保存的最新模型参数
+latest.pdopt     # 默认保存的最新模型的优化器相关参数
+```
+其中，best_accuracy是保存的精度最高的模型参数，可以直接使用该模型评估。
+
+模型评估命令如下：
+```
+python3 tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+训练的教师模型结构更大，精度更高，用于提升学生模型的精度。
+
+**提取教师模型参数**
+best_accuracy包含两个模型的参数，分别对应配置文件中的Student，Student2。提取Student的参数方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/dml_teacher.pdparams")
+```
+
+提取出来的模型参数可以用于模型进一步的finetune训练或者蒸馏训练。
+
+### 2.3 训练学生模型
+
+训练学生模型的配置文件是[ch_PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml)
+上一节训练得到的教师模型作为监督，采用CML方式训练得到轻量的学生模型。
+
+下载学生模型的ImageNet预训练模型：
+```
+# 下载MobileNetV3的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+```
+
+**启动训练**
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，
+模型评估命令如下：
+```
+python3 tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+best_accuracy包含三个模型的参数，分别对应配置文件中的Student，Student2，Teacher。提取Student参数的方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/cml_student.pdparams")
+```
+
+提取出来的Student的参数可用于模型部署或者做进一步的finetune训练。
+
+
+
+<a name="3"></a>
+## 3. 基于PPOCRv3检测finetune训练
+
+本节介绍如何使用PPOCRv3检测模型在其他场景上的finetune训练。
+
+finetune训练适用于三种场景：
+- 基于CML蒸馏方法的finetune训练，适用于教师模型在使用场景上精度高于PPOCRv3检测模型，且希望得到一个轻量检测模型。
+- 基于PPOCRv3轻量检测模型的finetune训练，无需训练教师模型，希望在PPOCRv3检测模型基础上提升使用场景上的精度。
+- 基于DML蒸馏方法的finetune训练，适用于采用DML方法进一步提升精度的场景。
+
+
+**基于CML蒸馏方法的finetune训练**
+
+下载PPOCRv3训练模型：
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams包含CML配置文件中Student、Student2、Teacher模型的参数。
+
+启动训练：
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy  \
+       Global.save_model_dir=./output/
+```
+
+**基于PPOCRv3轻量检测模型的finetune训练**
+
+
+下载PPOCRv3训练模型，并提取Student结构的模型参数：
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+提取Student参数的方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./student.pdparams")
+```
+
+使用配置文件[ch_PP-OCRv3_det_student.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml)训练。
+
+**启动训练**
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+```
+
+
+**基于DML蒸馏方法的finetune训练**
+
+以ch_PP-OCRv3_det_distill_train中的Teacher模型为例，首先提取Teacher结构的参数，方法如下：
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Teacher."):]: all_params[key] for key in all_params if "Teacher." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./teacher.pdparams")
+```
+
+**启动训练**
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+```
+
+

doc/doc_ch/algorithm_overview.md

@@ -89,6 +89,7 @@
 |ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
 |ABINet|Resnet45| 90.75% | rec_r45_abinet | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
+
 <a name="2"></a>
 
 ## 2. 端到端算法

doc/doc_ch/application.md

+# 场景应用
+
+PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
+
+> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
+
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
+
+## 通用
+
+| 类别                   | 亮点     | 类别       | 亮点         |
+| ---------------------- | -------- | ---------- | ------------ |
+| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
+
+## 制造
+
+| 类别           | 亮点                           | 类别           | 亮点                 |
+| -------------- | ------------------------------ | -------------- | -------------------- |
+| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
+| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
+| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
+
+## 金融
+
+| 类别           | 亮点                     | 类别         | 亮点                  |
+| -------------- | ------------------------ | ------------ | --------------------- |
+| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+
+## 交通
+
+| 类别              | 亮点                           | 类别       | 亮点     |
+| ----------------- | ------------------------------ | ---------- | -------- |
+| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file

doc/doc_en/algorithm_overview_en.md

@@ -88,6 +88,7 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
 |ABINet|Resnet45| 90.75% | rec_r45_abinet | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
+
 <a name="2"></a>
 
 ## 2. End-to-end Algorithms

ppocr/data/imaug/__init__.py

@@ -23,6 +23,7 @@ from .random_crop_data import EastRandomCropData, RandomCropImgMask
 from .make_pse_gt import MakePseGt
 
 
+
 from .rec_img_aug import BaseDataAugmentation, RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
                          SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
                          ABINetRecResizeImg, SVTRRecResizeImg, ABINetRecAug

ppocr/data/imaug/fce_targets.py

@@ -107,17 +107,20 @@ class FCENetTargets:
         for i in range(1, n):
             current_line_len = i * delta_length
 
-            while current_line_len >= length_cumsum[current_edge_ind + 1]:
+            while current_edge_ind + 1 < len(length_cumsum) and current_line_len >= length_cumsum[current_edge_ind + 1]:
                 current_edge_ind += 1
+
             current_edge_end_shift = current_line_len - length_cumsum[
                 current_edge_ind]
+
+            if current_edge_ind >= len(length_list):
+                break
             end_shift_ratio = current_edge_end_shift / length_list[
                 current_edge_ind]
             current_point = line[current_edge_ind] + (line[current_edge_ind + 1]
                                                       - line[current_edge_ind]
                                                       ) * end_shift_ratio
             resampled_line.append(current_point)
-
         resampled_line.append(line[-1])
         resampled_line = np.array(resampled_line)
 
@@ -328,6 +331,8 @@ class FCENetTargets:
             resampled_top_line, resampled_bot_line = self.resample_sidelines(
                 top_line, bot_line, self.resample_step)
             resampled_bot_line = resampled_bot_line[::-1]
+            if len(resampled_top_line) != len(resampled_bot_line):
+                continue
             center_line = (resampled_top_line + resampled_bot_line) / 2
 
             line_head_shrink_len = norm(resampled_top_line[0] -

ppocr/data/imaug/label_ops.py

@@ -23,7 +23,6 @@ import string
 from shapely.geometry import LineString, Point, Polygon
 import json
 import copy
-
 from ppocr.utils.logging import get_logger
 
 
@@ -74,9 +73,10 @@ class DetLabelEncode(object):
         s = pts.sum(axis=1)
         rect[0] = pts[np.argmin(s)]
         rect[2] = pts[np.argmax(s)]
-        diff = np.diff(pts, axis=1)
-        rect[1] = pts[np.argmin(diff)]
-        rect[3] = pts[np.argmax(diff)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
         return rect
 
     def expand_points_num(self, boxes):
@@ -407,10 +407,10 @@ class KieLabelEncode(object):
             texts.append(ann['transcription'])
             text_ind = [self.dict[c] for c in text if c in self.dict]
             text_inds.append(text_ind)
-            if 'label' in anno.keys():
+            if 'label' in ann.keys():
                 labels.append(ann['label'])
-            elif 'key_cls' in anno.keys():
-                labels.append(anno['key_cls'])
+            elif 'key_cls' in ann.keys():
+                labels.append(ann['key_cls'])
             else:
                 raise ValueError(
                     "Cannot found 'key_cls' in ann.keys(), please check your training annotation."

ppocr/utils/save_load.py

@@ -177,9 +177,9 @@ def save_model(model,
             model.backbone.model.save_pretrained(model_prefix)
         metric_prefix = os.path.join(model_prefix, 'metric')
     # save metric and config
+    with open(metric_prefix + '.states', 'wb') as f:
+        pickle.dump(kwargs, f, protocol=2)
     if is_best:
-        with open(metric_prefix + '.states', 'wb') as f:
-            pickle.dump(kwargs, f, protocol=2)
         logger.info('save best model is to {}'.format(model_prefix))
     else:
         logger.info("save model in {}".format(model_prefix))

test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

-===========================train_params===========================
-model_name:ch_PP-OCRv2_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
-fpgm_export: 
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv2_det_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

-===========================train_params===========================
-model_name:ch_PP-OCRv2_rec_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
-Global.pretrained_model:pretrain_models/ch_PP-OCRv2_rec_train/best_accuracy
-train_model_name:latest
-train_infer_img_dir:./inference/rec_inference
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
-fpgm_export: null
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv2_rec_slim_quant_infer
-infer_export:null
-infer_quant:True
-inference:tools/infer/predict_rec.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1|6
---use_tensorrt:False
---precision:fp32
---rec_model_dir:
---image_dir:./inference/rec_inference
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,32,320]}]

test_tipc/configs/ch_PP-OCRv3_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

-===========================train_params===========================
-model_name:ch_PP-OCRv3_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o 
-fpgm_export: 
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv3_det_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

-===========================train_params===========================
-model_name:ch_ppocr_mobile_v2.0_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
-fpgm_export:null
-distill_export:null
-export1:null
-export2:null
-inference_dir:null
-train_model:./inference/ch_ppocr_mobile_v2.0_det_prune_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]

test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

-===========================train_params===========================
-model_name:ch_ppocr_mobile_v2.0_rec_PACT
-python:python3.7
-gpu_list:0
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
-Global.checkpoints:null
-train_model_name:latest
-train_infer_img_dir:./train_data/ic15_data/test/word_1.png
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o 
-fpgm_export:null
-distill_export:null
-export1:null
-export2:null
-inference_dir:null
-infer_model:./inference/ch_ppocr_mobile_v2.0_rec_slim_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_image_shape="3,32,100"
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1|6
---use_tensorrt:False
---precision:fp32
---rec_model_dir:
---image_dir:./inference/rec_inference
---save_log_path:./test/output/
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,32,320]}]

test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt → test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt

 ===========================train_params===========================
-model_name:en_table_structure_PACT
+model_name:en_table_structure
 python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
 Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
 Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
@@ -12,9 +12,9 @@ train_model_name:latest
 train_infer_img_dir:./ppstructure/docs/table/table.jpg
 null:null
 ##
-trainer:pact_train
-norm_train:null 
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+pact_train:null
 fpgm_train:null
 distill_train:null
 null:null
@@ -27,8 +27,8 @@ null:null
 ===========================infer_params===========================
 Global.save_inference_dir:./output/
 Global.checkpoints:
-norm_export:null 
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+norm_export:tools/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+quant_export:
 fpgm_export: 
 distill_export:null
 export1:null
@@ -36,9 +36,9 @@ export2:null
 ##
 infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
 infer_export:null
-infer_quant:True
+infer_quant:False
 inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1

test_tipc/prepare.sh

@@ -100,7 +100,8 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
         wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar --no-check-certificate
-        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
+        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && cd ../
+        cd ./pretrain_models && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
     fi
     if [ ${model_name} == "det_mv3_db_v2_0" ]; then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate
@@ -353,7 +354,7 @@ elif [ ${MODE} = "whole_infer" ];then
     fi
 fi
 
-if [ ${MODE} = "klquant_whole_infer" ]; then
+if [[ ${model_name} =~ "KL" ]]; then
     wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar --no-check-certificate
     cd ./train_data/ && tar xf icdar2015_lite.tar && rm -rf ./icdar2015 && ln -s ./icdar2015_lite ./icdar2015 && cd ../
     if [ ${model_name} = "ch_ppocr_mobile_v2.0_det_KL" ]; then

test_tipc/test_paddle2onnx.sh

@@ -62,7 +62,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -73,7 +74,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -85,7 +87,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"     
@@ -97,7 +100,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"

test_tipc/test_ptq_inference_python.sh

+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of [''whole_infer']
+MODE=$2
+
+IFS=$'\n'
+# parser klquant_infer params
+
+dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
+lines=(${dataline})
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+export_weight=$(func_parser_key "${lines[3]}")
+save_infer_key=$(func_parser_key "${lines[4]}")
+# parser inference model 
+infer_model_dir_list=$(func_parser_value "${lines[5]}")
+infer_export_list=$(func_parser_value "${lines[6]}")
+infer_is_quant=$(func_parser_value "${lines[7]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[8]}")
+use_gpu_key=$(func_parser_key "${lines[9]}")
+use_gpu_list=$(func_parser_value "${lines[9]}")
+use_mkldnn_key=$(func_parser_key "${lines[10]}")
+use_mkldnn_list=$(func_parser_value "${lines[10]}")
+cpu_threads_key=$(func_parser_key "${lines[11]}")
+cpu_threads_list=$(func_parser_value "${lines[11]}")
+batch_size_key=$(func_parser_key "${lines[12]}")
+batch_size_list=$(func_parser_value "${lines[12]}")
+use_trt_key=$(func_parser_key "${lines[13]}")
+use_trt_list=$(func_parser_value "${lines[13]}")
+precision_key=$(func_parser_key "${lines[14]}")
+precision_list=$(func_parser_value "${lines[14]}")
+infer_model_key=$(func_parser_key "${lines[15]}")
+image_dir_key=$(func_parser_key "${lines[16]}")
+infer_img_dir=$(func_parser_value "${lines[16]}")
+save_log_key=$(func_parser_key "${lines[17]}")
+save_log_value=$(func_parser_value "${lines[17]}")
+benchmark_key=$(func_parser_key "${lines[18]}")
+benchmark_value=$(func_parser_value "${lines[18]}")
+infer_key1=$(func_parser_key "${lines[19]}")
+infer_value1=$(func_parser_value "${lines[19]}")
+
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                        continue
+                    fi # skip when quant model inference but precision is not int8
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then
+            save_infer_dir="${infer_model}_klquant"
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+            echo ${infer_run_exports[Count]} 
+            echo $export_cmd
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant="True"
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
+        Count=$(($Count + 1))
+    done
+fi
+

test_tipc/test_serving_infer_cpp.sh

@@ -70,7 +70,8 @@ function func_serving(){
     set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
     set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
     python_list=(${python_list})
-    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+    trans_det_log="${LOG_PATH}/cpp_trans_model_det.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
     eval $trans_model_cmd
     cp "deploy/pdserving/serving_client_conf.prototxt" ${det_serving_client_value}
     # trans rec
@@ -78,7 +79,8 @@ function func_serving(){
     set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
     set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
     python_list=(${python_list})
-    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+    trans_rec_log="${LOG_PATH}/cpp_trans_model_rec.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
     eval $trans_model_cmd
     last_status=${PIPESTATUS[0]}
     status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -88,22 +90,25 @@ function func_serving(){
     # cpp serving
     for gpu_id in ${gpu_value[*]}; do
         if [ ${gpu_id} = "null" ]; then
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > serving_log_cpu.log &"
+            server_log_path="${LOG_PATH}/cpp_server_cpu.log"
+            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > ${server_log_path} 2>&1 "
             eval $web_service_cpp_cmd
             last_status=${PIPESTATUS[0]}
             status_check $last_status "${web_service_cpp_cmd}" "${status_log}" "${model_name}"
             sleep 5s
-            _save_log_path="${LOG_PATH}/server_infer_cpp_cpu.log"
+            _save_log_path="${LOG_PATH}/cpp_client_cpu.log"
             cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
             eval $cpp_client_cmd
             last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}"
             status_check $last_status "${cpp_client_cmd}" "${status_log}" "${model_name}"
             ps ux | grep -i ${port_value} | awk '{print $2}' | xargs kill -s 9
         else
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > serving_log_gpu.log &"
+            server_log_path="${LOG_PATH}/cpp_server_gpu.log"
+            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > ${server_log_path} 2>&1 "
             eval $web_service_cpp_cmd
             sleep 5s
-            _save_log_path="${LOG_PATH}/server_infer_cpp_gpu.log"
+            _save_log_path="${LOG_PATH}/cpp_client_gpu.log"
             cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
             eval $cpp_client_cmd
             last_status=${PIPESTATUS[0]}

test_tipc/test_serving_infer_python.sh

@@ -77,14 +77,16 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         # trans rec
         set_dirname=$(func_set_params "--dirname" "${rec_infer_model_dir_value}")
         set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
     elif [[ ${model_name} =~ "det" ]]; then
         # trans det
@@ -92,7 +94,8 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
     elif [[ ${model_name} =~ "rec" ]]; then
         # trans rec
@@ -100,7 +103,8 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
     fi
     set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
@@ -115,29 +119,30 @@ function func_serving(){
             for use_mkldnn in ${web_use_mkldnn_list[*]}; do
                 for threads in ${web_cpu_threads_list[*]}; do
                     set_cpu_threads=$(func_set_params "${web_cpu_threads_key}" "${threads}")
+                    server_log_path="${LOG_PATH}/python_server_cpu_usemkldnn_${use_mkldnn}_threads_${threads}.log"
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     fi
                     sleep 2s
                     for pipeline in ${pipeline_py[*]}; do
-                        _save_log_path="${LOG_PATH}/server_infer_cpu_${pipeline%_client*}_usemkldnn_${use_mkldnn}_threads_${threads}_batchsize_1.log"
+                        _save_log_path="${LOG_PATH}/python_client_cpu_${pipeline%_client*}_usemkldnn_${use_mkldnn}_threads_${threads}_batchsize_1.log"
                         pipeline_cmd="${python} ${pipeline} ${set_image_dir} > ${_save_log_path} 2>&1 "
                         eval $pipeline_cmd
                         last_status=${PIPESTATUS[0]}
@@ -151,6 +156,7 @@ function func_serving(){
         elif [ ${use_gpu} = "gpu" ]; then
             for use_trt in ${web_use_trt_list[*]}; do
                 for precision in ${web_precision_list[*]}; do
+                    server_log_path="${LOG_PATH}/python_server_gpu_usetrt_${use_trt}_precision_${precision}.log"
                     if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
                         continue
                     fi
@@ -168,26 +174,26 @@ function func_serving(){
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     fi
                     sleep 2s
                     for pipeline in ${pipeline_py[*]}; do
-                        _save_log_path="${LOG_PATH}/server_infer_gpu_${pipeline%_client*}_usetrt_${use_trt}_precision_${precision}_batchsize_1.log"
+                        _save_log_path="${LOG_PATH}/python_client_gpu_${pipeline%_client*}_usetrt_${use_trt}_precision_${precision}_batchsize_1.log"
                         pipeline_cmd="${python} ${pipeline} ${set_image_dir}> ${_save_log_path} 2>&1"
                         eval $pipeline_cmd
                         last_status=${PIPESTATUS[0]}

test_tipc/test_train_inference_python.sh

@@ -2,7 +2,7 @@
 source test_tipc/common_func.sh
 
 FILENAME=$1
-# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer', 'klquant_whole_infer']
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer']
 MODE=$2
 
 dataline=$(awk 'NR==1, NR==51{print}'  $FILENAME)
@@ -88,43 +88,6 @@ benchmark_value=$(func_parser_value "${lines[49]}")
 infer_key1=$(func_parser_key "${lines[50]}")
 infer_value1=$(func_parser_value "${lines[50]}")
 
-# parser klquant_infer
-if [ ${MODE} = "klquant_whole_infer" ]; then
-    dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
-    lines=(${dataline})
-    model_name=$(func_parser_value "${lines[1]}")
-    python=$(func_parser_value "${lines[2]}")
-    export_weight=$(func_parser_key "${lines[3]}")
-    save_infer_key=$(func_parser_key "${lines[4]}")
-    # parser inference model 
-    infer_model_dir_list=$(func_parser_value "${lines[5]}")
-    infer_export_list=$(func_parser_value "${lines[6]}")
-    infer_is_quant=$(func_parser_value "${lines[7]}")
-    # parser inference 
-    inference_py=$(func_parser_value "${lines[8]}")
-    use_gpu_key=$(func_parser_key "${lines[9]}")
-    use_gpu_list=$(func_parser_value "${lines[9]}")
-    use_mkldnn_key=$(func_parser_key "${lines[10]}")
-    use_mkldnn_list=$(func_parser_value "${lines[10]}")
-    cpu_threads_key=$(func_parser_key "${lines[11]}")
-    cpu_threads_list=$(func_parser_value "${lines[11]}")
-    batch_size_key=$(func_parser_key "${lines[12]}")
-    batch_size_list=$(func_parser_value "${lines[12]}")
-    use_trt_key=$(func_parser_key "${lines[13]}")
-    use_trt_list=$(func_parser_value "${lines[13]}")
-    precision_key=$(func_parser_key "${lines[14]}")
-    precision_list=$(func_parser_value "${lines[14]}")
-    infer_model_key=$(func_parser_key "${lines[15]}")
-    image_dir_key=$(func_parser_key "${lines[16]}")
-    infer_img_dir=$(func_parser_value "${lines[16]}")
-    save_log_key=$(func_parser_key "${lines[17]}")
-    save_log_value=$(func_parser_value "${lines[17]}")
-    benchmark_key=$(func_parser_key "${lines[18]}")
-    benchmark_value=$(func_parser_value "${lines[18]}")
-    infer_key1=$(func_parser_key "${lines[19]}")
-    infer_value1=$(func_parser_value "${lines[19]}")
-fi
-
 LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
 mkdir -p ${LOG_PATH}
 status_log="${LOG_PATH}/results_python.log"
@@ -211,7 +174,7 @@ function func_inference(){
     done
 }
 
-if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
+if [ ${MODE} = "whole_infer" ]; then
     GPUID=$3
     if [ ${#GPUID} -le 0 ];then
         env=" "
@@ -226,16 +189,12 @@ if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
     infer_quant_flag=(${infer_is_quant})
     for infer_model in ${infer_model_dir_list[*]}; do
         # run export
-        if [ ${infer_run_exports[Count]} != "null" ];then
-            if [ ${MODE} = "klquant_whole_infer" ]; then
-                save_infer_dir="${infer_model}_klquant"
-            fi
-            if [ ${MODE} = "whole_infer" ]; then
-                save_infer_dir="${infer_model}"
-            fi
+        if [ ${infer_run_exports[Count]} != "null" ];then 
+            save_infer_dir="${infer_model}"
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key}"
+            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
             echo ${infer_run_exports[Count]} 
             echo $export_cmd
             eval $export_cmd
@@ -246,9 +205,6 @@ if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
         fi
         #run inference
         is_quant=${infer_quant_flag[Count]}
-        if [ ${MODE} = "klquant_whole_infer" ]; then
-            is_quant="True"
-        fi
         func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
         Count=$(($Count + 1))
     done
@@ -347,7 +303,8 @@ else
                 if [ ${eval_py} != "null" ]; then
                     eval ${env}
                     set_eval_params1=$(func_set_params "${eval_key1}" "${eval_value1}")
-                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1}" 
+                    eval_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_eval.log"
+                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1} > ${eval_log_path} 2>&1 " 
                     eval $eval_cmd
                     status_check $? "${eval_cmd}" "${status_log}" "${model_name}"
                 fi
@@ -355,9 +312,10 @@ else
                 if [ ${run_export} != "null" ]; then 
                     # run export model
                     save_infer_path="${save_log}"
+                    export_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_export.log"
                     set_export_weight=$(func_set_params "${export_weight}" "${save_log}/${train_model_name}")
                     set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_path}")
-                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key}"
+                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
                     eval $export_cmd
                     status_check $? "${export_cmd}" "${status_log}" "${model_name}"
 

tools/infer/predict_det.py

@@ -154,9 +154,10 @@ class TextDetector(object):
         s = pts.sum(axis=1)
         rect[0] = pts[np.argmin(s)]
         rect[2] = pts[np.argmax(s)]
-        diff = np.diff(pts, axis=1)
-        rect[1] = pts[np.argmin(diff)]
-        rect[3] = pts[np.argmax(diff)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
         return rect
 
     def clip_det_res(self, points, img_height, img_width):