# TyStudy **Repository Path**: tommycloud/TyStudy ## Basic Information - **Project Name**: TyStudy - **Description**: 前沿技术的研究与学习，会放一些示例在这里 - **Primary Language**: Java - **License**: Apache-2.0 - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 1 - **Forks**: 0 - **Created**: 2022-06-07 - **Last Updated**: 2025-08-28 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # TyStudy ## 介绍 前沿技术的研究与学习，会放一些示例在这里 - 机器学习、深度学习、神经网络等相关Demo - AI相关的测试Case - DJL Test Case - Paddle FastDeploy Test Case ## BoofCV **BoofCV** 是一个纯 Java 的全新实时计算机视觉库，其易于使用且具有非常高的性能。功能涵盖一系列Subjects、low-level图像处理、相机校准、特征检测/跟踪、运动结构、基准检测和识别。 [https://boofcv.org/](https://boofcv.org) ## DJL (Deep Java Library) `DJL (Deep Java Library)` 是一个用于深度学习的开源、高级、与引擎无关的 Java 框架。 亚马逊（Amazon）在2019 re:Invent 上宣布推出Deep Java Library (DJL)。通过DJL，用户可以使用Java来训练和部署深度学习模型。 ## FastDeploy FastDeploy面向AI模型产业落地，是一款全场景、易用灵活、极致高效的AI推理部署工具，支持云边端部署。支持160+开箱即用的AI模型在8大类常见硬件上的部署能力。套件包括目标检测、字符识别（OCR）、人脸、人像扣图、多目标跟踪系统、NLP、Stable Diffusion文图生成、TTS 等几十种任务场景。 ## DJL Paddle FastDeploy 示例 百度飞浆 `FastDeploy` 产品的示例，基于`DJL Python Engine`实现。 **准备工作：** - 请参考百度飞浆`FastDeploy`官网（[https://www.paddlepaddle.org.cn/fastdeploy](https://www.paddlepaddle.org.cn/fastdeploy)），安装FastDeploy； - 配置环境变量 **PYTHON_EXECUTABLE** （Python执行文件的完整路径） > 如：PYTHON_EXECUTABLE = C:\DevTools\miniconda\envs\cuda117\python.exe > 注：Python环境建议用 **Anoconda** 或 **miniConda** 搭建。 - Clone 本项目，运行 [src/main/java/com/ty/test/ai/cv/PPYoloEPlusModelTest.java](https://gitee.com/tommycloud/TyStudy/blob/master/src/main/java/com/ty/test/ai/cv/PPYoloEPlusModelTest.java) 查看效果。 ```sh # 成功运行输出如下： 2024-05-03 22:21:26.493 [W-7380-ppyoloEplus-stdout] INFO ai.djl.python.engine.PyProcess - W-7380-ppyoloEplus-stdout: [INFO] fastdeploy/runtime/runtime.cc(273)::fastdeploy::Runtime::CreatePaddleBackend Runtime initialized with Backend::PDINFER in Device::GPU. 2024-05-03 22:21:26.500 [W-7380-ppyoloEplus-stdout] INFO ai.djl.python.engine.PyProcess - W-7380-ppyoloEplus-stdout: PPYoloE Plus Model Loaded, spend time 6 second. 2024-05-03 22:21:27.050 [main] INFO c.ty.test.ai.cv.PPYoloEPlusModelTest - 推理结束：[ {"class": "风筝", "probability": 0.97197, "bounds": {"x"=161.763, "y"=81.466, "width"=455.528, "height"=263.177}} {"class": "人", "probability": 0.90937, "bounds": {"x"=163.771, "y"=81.891, "width"=34.823, "height"=85.555}} {"class": "人", "probability": 0.90780, "bounds": {"x"=267.230, "y"=89.545, "width"=31.428, "height"=79.574}} {"class": "人", "probability": 0.89662, "bounds": {"x"=581.122, "y"=113.050, "width"=31.693, "height"=85.292}} {"class": "人", "probability": 0.88805, "bounds": {"x"=414.957, "y"=89.496, "width"=92.144, "height"=195.913}} {"class": "人", "probability": 0.80455, "bounds": {"x"=363.936, "y"=58.531, "width"=17.751, "height"=55.904}} {"class": "人", "probability": 0.79903, "bounds": {"x"=503.756, "y"=113.386, "width"=91.321, "height"=157.999}} {"class": "椅子", "probability": 0.79050, "bounds": {"x"=75.447, "y"=121.782, "width"=28.761, "height"=34.099}} {"class": "人", "probability": 0.78817, "bounds": {"x"=104.796, "y"=45.746, "width"=22.949, "height"=47.837}} {"class": "人", "probability": 0.76899, "bounds": {"x"=327.770, "y"=39.126, "width"=19.448, "height"=41.508}} {"class": "人", "probability": 0.72002, "bounds": {"x"=348.658, "y"=44.040, "width"=19.439, "height"=54.283}} {"class": "人", "probability": 0.69320, "bounds": {"x"=378.745, "y"=40.857, "width"=16.668, "height"=43.075}} {"class": "人", "probability": 0.62536, "bounds": {"x"=26.247, "y"=118.027, "width"=35.093, "height"=35.516}} {"class": "椅子", "probability": 0.61076, "bounds": {"x"=98.472, "y"=130.829, "width"=18.842, "height"=25.674}} {"class": "人", "probability": 0.52820, "bounds": {"x"=169.729, "y"=47.327, "width"=8.937, "height"=13.904}} {"class": "背包", "probability": 0.50318, "bounds": {"x"=65.103, "y"=134.649, "width"=19.887, "height"=19.887}} ] ``` ![test1.jpg预测结果](https://gitee.com/tommycloud/TyStudy/raw/master/src/main/resources/test/test1_visualized_result.jpg) ## DJL Paddle ONNX 示例 使用 `Paddle2onnx`工具将Paddle模型转换为ONNX模型，基于`DJL ONNX Engine`实现 AI 推理。 - Clone 本项目，运行 [src/main/java/com/ty/test/ai/cv/PPYoloEPlusONNXModelTest.java](https://gitee.com/tommycloud/TyStudy/blob/master/src/main/java/com/ty/test/ai/cv/PPYoloEPlusONNXModelTest.java) 查看效果。 ```sh # 成功运行输出如下： 2024-05-15 00:24:33.943 [main] DEBUG ai.djl.engine.Engine - Found default engine: PyTorch 2024-05-15 00:24:37.986 [main] DEBUG ai.djl.repository.zoo.ModelZoo - Checking ModelLoader: ai.djl.localmodelzoo:ppyoloE_plus UNDEFINED [ ai.djl.localmodelzoo/ppyoloE_plus/ppyoloE_plus {} ] 2024-05-15 00:24:38.001 [main] DEBUG ai.djl.repository.MRL - Preparing artifact: file:/opt/onnx_models/ppyoloE_plus/, ai.djl.localmodelzoo/ppyoloE_plus/ppyoloE_plus {} 2024-05-15 00:24:38.002 [main] DEBUG ai.djl.repository.SimpleRepository - Skip prepare for local repository. Loading: 100% |========================================| 2024-05-15 00:24:40.924 [main] INFO c.t.a.c.p.m.p.PPYoloEPlusONNXModel - PPYoloE Plus ONNX模型加载完毕 2024-05-15 00:24:40.925 [main] INFO c.t.t.ai.cv.PPYoloEPlusONNXModelTest - 开始推理：src\main\resources\test\test1.jpg 2024-05-15 00:24:40.925 [main] DEBUG c.t.a.c.p.t.PaddleDetectionTranslator - Process Input 数据预处理... 2024-05-15 00:24:40.991 [main] DEBUG c.t.a.c.p.t.PaddleDetectionTranslator - Process Input 数据预处理完成，耗时：66ms 2024-05-15 00:24:41.695 [main] DEBUG c.t.a.c.p.t.PaddleDetectionTranslator - Process Output 数据后处理... 2024-05-15 00:24:41.704 [main] DEBUG c.t.a.c.p.t.PaddleDetectionTranslator - Process Output 数据后处理完成，耗时：8ms 2024-05-15 00:24:42.259 [main] TRACE c.t.a.c.p.t.PaddleDetectionTranslator - 目标检测结果集（300条） label | score | x_min | y_min | x_max | y_max | ---------------------------------------------------------------------------------------------------------------------------- 33 | 0.9723131060600281 | 161.77688598632812 | 81.42156982421875 | 616.650146484375 | 344.8529052734375 | 0 | 0.9130071401596069 | 163.8280487060547 | 81.87583923339844 | 198.5448455810547 | 167.5631103515625 | 0 | 0.90949547290802 | 267.20965576171875 | 89.6672592163086 | 298.622314453125 | 169.03573608398438 ... | ... | ... | ... | ... | ... 2024-05-15 00:24:42.329 [main] INFO c.t.t.ai.cv.PPYoloEPlusONNXModelTest - 推理结束 src\main\resources\test\test1.jpg 目标检测数量：15 ``` ![test1.jpg预测结果](https://gitee.com/tommycloud/TyStudy/raw/master/src/main/resources/test/test1_visualized_result.jpg) ## 使用的开源算法 [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection) ## 预训练模型 - PPYoloE Plus 预训练模型下载地址 > https://pan.baidu.com/s/1EFbAoHujknZgrtlzsA4t0Q?pwd=a42d > - PPYoloE Plus ONNX 预训练模型下载地址 > https://pan.baidu.com/s/1aK1vNSd-I1l7TB_e0_Df6w?pwd=sdty