diff --git a/examples/vision/keypointdetection/tiny_pose/rknpu2/README.md b/examples/vision/keypointdetection/tiny_pose/rknpu2/README.md index f7f270c0a..100234ac8 100644 --- a/examples/vision/keypointdetection/tiny_pose/rknpu2/README.md +++ b/examples/vision/keypointdetection/tiny_pose/rknpu2/README.md @@ -51,5 +51,5 @@ python tools/rknpu2/export.py --config_path tools/rknpu2/config/PP_TinyPose_256x ## 详细部署文档 - [模型详细介绍](../README_CN.md) -- [Python部署](python) -- [C++部署](cpp) \ No newline at end of file +- [Python部署](./python) +- [C++部署](./cpp) \ No newline at end of file diff --git a/examples/vision/keypointdetection/tiny_pose/rknpu2/python/README_CN.md b/examples/vision/keypointdetection/tiny_pose/rknpu2/python/README_CN.md new file mode 100644 index 000000000..1a0f37d0b --- /dev/null +++ b/examples/vision/keypointdetection/tiny_pose/rknpu2/python/README_CN.md @@ -0,0 +1,70 @@ +[English](README.md) | 简体中文 +# PP-TinyPose 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/download_prebuilt_libraries.md) +- 2. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md) + +本目录下提供`pptinypose_infer.py`快速完成PP-TinyPose在NPU加速部署的`单图单人关键点检测`示例。执行如下脚本即可完成 + +>> **注意**: PP-Tinypose单模型目前只支持单图单人关键点检测，因此输入的图片应只包含一个人或者进行过裁剪的图像。多人关键点检测请参考[PP-TinyPose Pipeline](../../../det_keypoint_unite/python/README.md) + +```bash +# 下载PP-TinyPose模型文件和测试图片 +wget https://bj.bcebos.com/paddlehub/fastdeploy/hrnet_demo.jpg + +# CPU推理 +python pptinypose_infer.py --tinypose_model_dir PP_TinyPose_256x192_infer --image hrnet_demo.jpg +``` + +运行完成可视化结果如下图所示 +
+ +
+ +## PP-TinyPose Python接口 + +```python +fd.vision.keypointdetection.PPTinyPose(model_file, params_file, config_file, runtime_option=None, model_format=ModelFormat.PADDLE) +``` + +PP-TinyPose模型加载和初始化，其中model_file, params_file以及config_file为训练模型导出的Paddle inference文件，具体请参考其文档说明[模型导出](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/deploy/EXPORT_MODEL.md) + +**参数** + +> * **model_file**(str): 模型文件路径 +> * **params_file**(str): 参数文件路径 +> * **config_file**(str): 推理部署配置文件 +> * **runtime_option**(RuntimeOption): 后端推理配置，默认为None，即采用默认配置 +> * **model_format**(ModelFormat): 模型格式，默认为Paddle格式 + +### predict函数 + +> ```python +> PPTinyPose.predict(input_image) +> ``` +> +> 模型预测结口，输入图像直接输出检测结果。 +> +> **参数** +> +> > * **input_image**(np.ndarray): 输入数据，注意需为HWC，BGR格式 + +> **返回** +> +> > 返回`fastdeploy.vision.KeyPointDetectionResult`结构体，结构体说明参考文档[视觉模型预测结果](../../../../../docs/api/vision_results/) + +### 类成员属性 +#### 后处理参数 +用户可按照自己的实际需求，修改下列后处理参数，从而影响最终的推理和部署效果 + +> > * **use_dark**(bool): 是否使用DARK进行后处理[参考论文](https://arxiv.org/abs/1910.06278) + + +## 其它文档 + +- [PP-TinyPose 模型介绍](..) +- [PP-TinyPose C++部署](../cpp) +- [模型预测结果说明](../../../../../docs/api/vision_results/) +- [如何切换模型推理后端引擎](../../../../../docs/cn/faq/how_to_change_backend.md) diff --git a/examples/vision/keypointdetection/tiny_pose/rknpu2/python/pptinypose_infer.py b/examples/vision/keypointdetection/tiny_pose/rknpu2/python/pptinypose_infer.py new file mode 100755 index 000000000..a9e75dfa7 --- /dev/null +++ b/examples/vision/keypointdetection/tiny_pose/rknpu2/python/pptinypose_infer.py @@ -0,0 +1,50 @@ +import fastdeploy as fd +import cv2 +import os + + +def parse_arguments(): + import argparse + import ast + parser = argparse.ArgumentParser() + parser.add_argument( + "--tinypose_model_dir", + required=True, + help="path of paddletinypose model directory") + parser.add_argument( + "--image", required=True, help="path of test image file.") + return parser.parse_args() + + +def build_tinypose_option(args): + option = fd.RuntimeOption() + option.use_rknpu() + return option + + +args = parse_arguments() + +tinypose_model_file = os.path.join(args.tinypose_model_dir, "PP_TinyPose_256x192_infer_rk3588_unquantized.rknn") +tinypose_params_file = os.path.join(args.tinypose_model_dir, "") +tinypose_config_file = os.path.join(args.tinypose_model_dir, "infer_cfg.yml") +# 配置runtime，加载模型 +runtime_option = build_tinypose_option(args) +tinypose_model = fd.vision.keypointdetection.PPTinyPose( + tinypose_model_file, + tinypose_params_file, + tinypose_config_file, + runtime_option=runtime_option, + model_format=fd.ModelFormat.RKNN) +tinypose_model.disable_normalize() +tinypose_model.disable_permute() + +# 预测图片检测结果 +im = cv2.imread(args.image) +tinypose_result = tinypose_model.predict(im) +print("Paddle TinyPose Result:\n", tinypose_result) + +# 预测结果可视化 +vis_im = fd.vision.vis_keypoint_detection( + im, tinypose_result, conf_threshold=0.5) +cv2.imwrite("visualized_result.jpg", vis_im) +print("TinyPose visualized result save in ./visualized_result.jpg") diff --git a/fastdeploy/vision/keypointdet/pptinypose/pptinypose.cc b/fastdeploy/vision/keypointdet/pptinypose/pptinypose.cc index f4b3bfc6c..1d1ad5c3f 100644 --- a/fastdeploy/vision/keypointdet/pptinypose/pptinypose.cc +++ b/fastdeploy/vision/keypointdet/pptinypose/pptinypose.cc @@ -139,18 +139,12 @@ bool PPTinyPose::Postprocess(std::vector& infer_result, "Only support batch = 1 in FastDeploy now."); result->Clear(); - std::cout << "Postprocess" << std::endl; - std::cout << "infer_result.size() is " << infer_result.size() << std::endl; if (infer_result.size() == 1) { FDTensor result_copy = infer_result[0]; - std::cout << "Reshape result_copy!" << std::endl; result_copy.Reshape({result_copy.shape[0], result_copy.shape[1], result_copy.shape[2] * result_copy.shape[3]}); - std::cout << "Resize infer_result!" << std::endl; infer_result.resize(2); - std::cout << "Do ArgMax!" << std::endl; function::ArgMax(result_copy,&infer_result[1],-1); - std::cout << "Done!" << std::endl; } // Calculate output length