add ocr, ppyoloe, picodet examples (#1076)

* add ocr examples

* add ppyoloe examples

add picodet examples

* remove /ScaleFactor in ppdet/postprocessor.cc
This commit is contained in:
Dantès
2023-01-10 16:34:26 +08:00
committed by GitHub
parent fc314f1696
commit de70e8366c
21 changed files with 922 additions and 13 deletions

View File

@@ -0,0 +1,38 @@
# PPOCRv3 Python部署示例
在部署前,需确认以下两个步骤
- 1. 软硬件环境满足要求,参考[FastDeploy环境要求](../../../../../../docs/cn/build_and_install/sophgo.md)
本目录下提供`infer.py`快速完成 PPOCRv3 在SOPHGO TPU上部署的示例。执行如下脚本即可完成
```bash
# 下载部署示例代码
git clone https://github.com/PaddlePaddle/FastDeploy.git
cd FastDeploy/examples/vision/ocr/PP-OCRv3/sophgo/python
# 下载图片
wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/doc/imgs/12.jpg
#下载字典文件
wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_keys_v1.txt
# 推理
python3 infer.py --det_model ocr_bmodel/ch_PP-OCRv3_det_1684x_f32.bmodel \
--cls_model ocr_bmodel/ch_ppocr_mobile_v2.0_cls_1684x_f32.bmodel \
--rec_model ocr_bmodel/ch_PP-OCRv3_rec_1684x_f32.bmodel \
--rec_label_file ../ppocr_keys_v1.txt \
--image ../12.jpg
# 运行完成后返回结果如下所示
det boxes: [[42,413],[483,391],[484,428],[43,450]]rec text: 上海斯格威铂尔大酒店 rec score:0.952958 cls label: 0 cls score: 1.000000
det boxes: [[187,456],[399,448],[400,480],[188,488]]rec text: 打浦路15号 rec score:0.897335 cls label: 0 cls score: 1.000000
det boxes: [[23,507],[513,488],[515,529],[24,548]]rec text: 绿洲仕格维花园公寓 rec score:0.994589 cls label: 0 cls score: 1.000000
det boxes: [[74,553],[427,542],[428,571],[75,582]]rec text: 打浦路252935号 rec score:0.900663 cls label: 0 cls score: 1.000000
可视化结果保存在sophgo_result.jpg中
```
## 其它文档
- [PPOCRv3 C++部署](../cpp)
- [转换 PPOCRv3 SOPHGO模型文档](../README.md)

View File

@@ -0,0 +1,116 @@
import fastdeploy as fd
import cv2
import os
def parse_arguments():
import argparse
import ast
parser = argparse.ArgumentParser()
parser.add_argument(
"--det_model", required=True, help="Path of Detection model of PPOCR.")
parser.add_argument(
"--cls_model",
required=True,
help="Path of Classification model of PPOCR.")
parser.add_argument(
"--rec_model",
required=True,
help="Path of Recognization model of PPOCR.")
parser.add_argument(
"--rec_label_file",
required=True,
help="Path of Recognization label of PPOCR.")
parser.add_argument(
"--image", type=str, required=True, help="Path of test image file.")
return parser.parse_args()
args = parse_arguments()
# 配置runtime加载模型
runtime_option = fd.RuntimeOption()
runtime_option.use_sophgo()
# Detection模型, 检测文字框
det_model_file = args.det_model
det_params_file = ""
# Classification模型方向分类可选
cls_model_file = args.cls_model
cls_params_file = ""
# Recognition模型文字识别模型
rec_model_file = args.rec_model
rec_params_file = ""
rec_label_file = args.rec_label_file
# PPOCR的cls和rec模型现在已经支持推理一个Batch的数据
# 定义下面两个变量后, 可用于设置trt输入shape, 并在PPOCR模型初始化后, 完成Batch推理设置
cls_batch_size = 1
rec_batch_size = 1
# 当使用TRT时分别给三个模型的runtime设置动态shape,并完成模型的创建.
# 注意: 需要在检测模型创建完成后,再设置分类模型的动态输入并创建分类模型, 识别模型同理.
# 如果用户想要自己改动检测模型的输入shape, 我们建议用户把检测模型的长和高设置为32的倍数.
det_option = runtime_option
det_option.set_trt_input_shape("x", [1, 3, 64, 64], [1, 3, 640, 640],
[1, 3, 960, 960])
# 用户可以把TRT引擎文件保存至本地
# det_option.set_trt_cache_file(args.det_model + "/det_trt_cache.trt")
det_model = fd.vision.ocr.DBDetector(
det_model_file,
det_params_file,
runtime_option=det_option,
model_format=fd.ModelFormat.SOPHGO)
cls_option = runtime_option
cls_option.set_trt_input_shape("x", [1, 3, 48, 10],
[cls_batch_size, 3, 48, 320],
[cls_batch_size, 3, 48, 1024])
# 用户可以把TRT引擎文件保存至本地
# cls_option.set_trt_cache_file(args.cls_model + "/cls_trt_cache.trt")
cls_model = fd.vision.ocr.Classifier(
cls_model_file,
cls_params_file,
runtime_option=cls_option,
model_format=fd.ModelFormat.SOPHGO)
rec_option = runtime_option
rec_option.set_trt_input_shape("x", [1, 3, 48, 10],
[rec_batch_size, 3, 48, 320],
[rec_batch_size, 3, 48, 2304])
# 用户可以把TRT引擎文件保存至本地
# rec_option.set_trt_cache_file(args.rec_model + "/rec_trt_cache.trt")
rec_model = fd.vision.ocr.Recognizer(
rec_model_file,
rec_params_file,
rec_label_file,
runtime_option=rec_option,
model_format=fd.ModelFormat.SOPHGO)
# 创建PP-OCR串联3个模型其中cls_model可选如无需求可设置为None
ppocr_v3 = fd.vision.ocr.PPOCRv3(
det_model=det_model, cls_model=cls_model, rec_model=rec_model)
# 需要使用下行代码, 来启用rec模型的静态shape推理这里rec模型的静态输入为[3, 48, 584]
rec_model.preprocessor.static_shape_infer = True
rec_model.preprocessor.rec_image_shape = [3, 48, 584]
# 给cls和rec模型设置推理时的batch size
# 此值能为-1, 和1到正无穷
# 当此值为-1时, cls和rec模型的batch size将默认和det模型检测出的框的数量相同
ppocr_v3.cls_batch_size = cls_batch_size
ppocr_v3.rec_batch_size = rec_batch_size
# 预测图片准备
im = cv2.imread(args.image)
#预测并打印结果
result = ppocr_v3.predict(im)
print(result)
# 可视化结果
vis_im = fd.vision.vis_ppocr(im, result)
cv2.imwrite("sophgo_result.jpg", vis_im)
print("Visualized result save in ./sophgo_result.jpg")