PaddleX

简介

PaddleX 3.0是飞桨精选模型的低代码开发工具，支持国内外多款主流硬件的模型训练和推理，覆盖工业、能源、金融、交通、教育等全行业，助力开发者产业实践落地。

通用图像分类	通用目标检测	通用语义分割	通用实例分割
<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/b302cd7e-e027-4ea6-86d0-8a4dd6d61f39" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/099e2b00-0bbe-4b20-9c5a-96b69e473bd2" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/02637f8c-f248-415b-89ab-1276505f198c" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/09f683b4-27df-4c24-b8a7-84da20fdd182" height="126px" width="180px">
通用OCR	通用表格识别	通用场景信息抽取	文档场景信息抽取
<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/1ef48536-48d4-484b-a6fb-0d6631ba2386" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/1e798e05-dee7-4b41-9cc4-6708b6014efa" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/04218629-4a7b-48ea-b815-977a05fbbb13" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/e3d97f4e-ab46-411c-8155-494c61492b0a" height="126px" width="180px">
时序预测	时序异常检测	时序分类	多模型融合时序预测
<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/6e897bf6-35fe-45e6-a040-e9a1a20cfdf2" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/c54c66cc-da4f-4631-877b-43b0fbb192a6" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/0ce925b2-3776-4dde-8ce0-5156d5a2476e" height="126px" width="180px">	<img src="https://github.com/PaddlePaddle/PaddleX/assets/142379845/0959d099-a17c-40bc-9c2b-13f4f5e24ddc" height="126px" width="180px">

📣 近期更新

🔥🔥《PaddleX图像分割产线赋能医疗领域筛查检测》，零代码开发实现恶性皮肤肿瘤早期筛查。跨学科合作提升医疗诊断效率与质量，打造智能医疗诊断系统，实现高效精准的图像分割效果。适用医学影像处理、工业检测、无人机航拍、智慧交通、安防监控等产业实际场景。8月29日（周四）19：00直播分享图像分割全流程解决方案，解密模型选型与快速调参技法。报名链接。
🔥 2024年6月27日，PaddleX 3.0 Beta 本地端正式发布，支持以低代码的方式在本地端使用多种主流硬件进行产线和模型开发。
🔥 2024年3月25日，PaddleX 3.0 云端发布，支持在AI Studio 星河社区以零代码的方式【创建产线】使用。

🌟 特性

PaddleX 3.0集成了飞桨生态的优势能力，覆盖7大场景任务，构建了16条模型产线，提供低代码开发模式，助力开发者在多种主流硬件上实现模型全流程开发。

Basic Model Pipeline (Rich Models, Comprehensive Scenarios): Carefully selected 68 high-quality PaddlePaddle models, covering image classification, object detection, image segmentation, OCR, text-image layout analysis, text-image information extraction, and time series analysis task scenarios.
Featured Model Pipeline (Significantly Improves Efficiency): Provides efficient solutions combining large and small models, semi-supervised learning with large models, and multi-model fusion.
Low-threshold Development Mode (Convenient Development and Deployment): Offers two development approaches: zero-code and low-code.
- Zero-code Development: Users submit backend training tasks through a graphical user interface (GUI), enabling both online and offline deployment, and supporting online service calls via API.
- Low-code Development: Implements full-process development of 16 model pipelines through a unified API interface, while supporting user-defined model process chaining.
Multi-hardware Local Support (Strong Compatibility): Supports various hardware including NVIDIA GPUs, Kunlun chips, Ascend, and Cambricon, for pure offline use.

⚡ Installation and Quick Start

Installation
Quick Start
- Model Pipeline Development Tools
- Single Model Development Tools

🛠️ PaddleX 3.0 涵盖的模型和模型产线

PaddleX 3.0 涵盖了16条产业级模型产线，其中9条基础产线可直接在本仓库离线使用，7条特色产线可在飞桨 AI Studio 星河社区上免费使用。

<table> <tr> <th>产线类型</th> <th>模型产线</th> <th>产线模块</th> <th>具体模型</th> </tr> <tr> <td>基础产线</td> <td>通用图像分类</td> <td>图像分类</td> <td>CLIP_vit_base_patch16_224 CLIP_vit_large_patch14_224<details> <summary>更多</summary> ConvNeXt_tiny MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x1_0 MobileNetV2_x1_5 MobileNetV2_x2_0 MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 MobileNetV3_small_x1_25 PP-HGNet_small PP-HGNetV2-B0 PP-HGNetV2-B4 PP-HGNetV2-B6 PP-LCNet_x0_25 PP-LCNet_x0_35 PP-LCNet_x0_5 PP-LCNet_x0_75 PP-LCNet_x1_0 PP-LCNet_x1_5 PP-LCNet_x2_0 PP-LCNet_x2_5 ResNet18 ResNet34 ResNet50 ResNet101 ResNet152 SwinTransformer_base_patch4_window7_224</details></td> </tr> <tr> <td>基础产线</td> <td>通用目标检测</td> <td>目标检测</td> <td>PicoDet-S PicoDet-L<details> <summary>更多</summary> PP-YOLOE_plus-S PP-YOLOE_plus-M PP-YOLOE_plus-L PP-YOLOE_plus-X RT-DETR-L RT-DETR-H RT-DETR-X RT-DETR-R18 RT-DETR-R50</details></td> </tr> <tr> <td>基础产线</td> <td>通用语义分割</td> <td>语义分割</td> <td>OCRNet_HRNet-W48 PP-LiteSeg-T<details> <summary>更多</summary> Deeplabv3-R50 Deeplabv3-R101 Deeplabv3_Plus-R50 Deeplabv3_Plus-R101</details></td> </tr> <tr> <td>基础产线</td> <td>通用实例分割</td> <td>实例分割</td> <td>Mask-RT-DETR-L Mask-RT-DETR-H</td> </tr> <tr> <td rowspan="2">基础产线</td> <td rowspan="2">通用OCR</td> <td>文本检测</td> <td>PP-OCRv4_mobile_det PP-OCRv4_server_det</td> </tr> <tr> <td>文本识别</td> <td>PP-OCRv4_mobile_rec PP-OCRv4_server_rec</td> </tr> <tr> <td rowspan="4">基础产线</td> <td rowspan="4">通用表格识别</td> <td>版面区域检测</td> <td>PicoDet layout_1x</td> </tr> <tr> <td>表格识别</td> <td>SLANet</td> </tr> <tr> <td>文本检测</td> <td>PP-OCRv4_mobile_det PP-OCRv4_server_det</td> </tr> <tr> <td>文本识别</td> <td>PP-OCRv4_mobile_rec PP-OCRv4_server_rec</td> </tr> <tr> <td>基础产线</td> <td>时序预测</td> <td>时序预测</td> <td>DLinear Nonstationary TiDE PatchTST TimesNet</td> </tr> <tr> <td>基础产线</td> <td>时序异常检测</td> <td>时序异常检测</td> <td>DLinear_ad Nonstationary_ad AutoEncoder_ad PatchTST_ad TimesNet_ad</td> </tr> <tr> <td>基础产线</td> <td>时序分类</td> <td>时序分类</td> <td>TimesNet_cls</td> </tr> <tr> <td>特色产线</td> <td>大模型半监督学习-图像分类</td> <td>大模型半监督学习-图像分类</td> <td>CLIP_vit_base_patch16_224 MobileNetV3_small_x1_0 <details><summary>更多</summary>PP-HGNet_small PP-HGNetV2-B0 PP-HGNetV2-B4 PP-HGNetV2-B6 PP-LCNet_x1_0 ResNet50 SwinTransformer_base_patch4_window7_224</details></td> </tr> <tr> <td>特色产线</td> <td>大模型半监督学习-目标检测</td> <td>大模型半监督学习-目标检测</td> <td>PicoDet-S PicoDet-L<details> <summary>更多</summary> PP-YOLOE plus-S PP-YOLOE_plus-L RT-DETR-H</details></td> </tr> <tr> <td rowspan="2">特色产线</td> <td rowspan="2">大模型半监督学习-OCR</td> <td>文本检测</td> <td>PP-OCRv4_mobile_det PP-OCRv4_server_det</td> </tr> <tr> <td>大模型半监督学习-文本识别</td> <td>PP-OCRv4_mobile_rec PP-OCRv4_server_rec</td> </tr> <tr> <td rowspan="3">特色产线</td> <td rowspan="3">通用场景信息抽取v2 (PP-ChatOCRv2-common)</td> <td>文本识别</td> <td>PP-OCRv4_mobile_rec PP-OCRv4_server_rec</td> </tr> <tr> <td>文本检测</td> <td>PP-OCRv4_mobile_det PP-OCRv4_server_det</td> </tr> <tr> <td>提示工程</td> <td>-</td> </tr> <tr> <td rowspan="5">特色产线</td> <td rowspan="5">文档场景信息抽取v2 (PP-ChatOCRv2-doc)</td> <td>版面分析</td> <td>PicoDet layout_1x</td> </tr> <tr> <td>文本检测</td> <td>PP-OCRv4_mobile_det PP-OCRv4_server_det</td> </tr> <tr> <td>文本识别</td> <td>PP-OCRv4_mobile_rec PP-OCRv4_server_rec</td> </tr> <tr> <td>表格识别</td> <td>SLANet</td> </tr> <tr> <td>提示工程</td> <td>-</td> </tr> <tr> <td>特色产线</td> <td>多模型融合时序预测v2 (PP-TSv2_forecast)</td> <td>时序预测</td> <td>多模型融合时序预测</td> </tr> <tr> <td>特色产线</td> <td>多模型融合时序异常检测v2 (PP-TSv2_anomaly)</td> <td>时序异常检测</td> <td>多模型融合时序异常检测</td> </tr> </table> - [模型流水线列表](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta/./docs/tutorials/pipelines/support_pipeline_list.md) - [单模型列表](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta/./docs/tutorials/models/support_model_list.md)