PaddlePaddle/PaddleX

Features

• Computer Vision Pipelines - Provides a toolkit of pre-built pipelines for image classification, detection, segmentation, and pose estimation.
• Pipeline Frameworks - Provides the core framework for building, deploying, and fine-tuning modular AI model pipelines.
• Object Detection - Identifies multiple object categories and their bounding-box locations within images.
• Computer Vision Toolkits - Offers pre-built pipelines for image classification, object detection, segmentation, and pose estimation.
• Document Information Extraction - Combines OCR, layout analysis, and LLMs to extract structured information from complex scanned documents.
• Detection Model Selections - Offers dozens of pre-trained detection models with configurable accuracy-speed trade-offs.
• NPU Inference Execution - Switches between GPU, NPU, and other hardware devices with a single parameter change for inference.
• Image Classification - Assigns images to predefined categories using trained deep-learning models.
• Image Text Translators - Converts printed, handwritten, or symbolic text in images into editable text using a multi-stage OCR pipeline.
• Multi-Backend GPU Inference Engines - Ships a multi-backend inference engine supporting Paddle Inference, TensorRT, OpenVINO, ONNX Runtime, and Ascend OM.
• Hardware-Agnostic Inference Switching - Changes the computing device for model inference by setting a single parameter, supporting GPUs, NPUs, and other accelerators.
• Model Serving & Deployment - Deploys PaddlePaddle model pipelines as REST APIs, edge device binaries, or high-performance inference services.
• Open-Vocabulary Object Detection - Identifies and locates objects in images using natural language prompts instead of predefined categories.
• Pipeline Executors - Provides a unified interface to run complete pre-trained model pipelines on input data via CLI or Python.
• Inference Pipelines - Provides a pipeline-based inference system for running AI models on input data.
• Time Series Forecasting - Ships a dedicated pipeline for forecasting, classifying, and detecting anomalies in time series data.
• OCR Pipelines - Provides an end-to-end pipeline for extracting text, tables, formulas, and layout from documents.
• CLI Executions - Provides a CLI command that runs the full OCR pipeline on images or PDFs.
• Model Fine-Tuning - Retrains pre-trained models on domain-specific datasets to improve accuracy for specialized tasks.
• Text Detection - Locates text areas in images with bounding boxes for downstream recognition.
• Text Recognition - Transcribes printed or handwritten text from image regions into machine-readable strings.
• Table Structure Detections - Implements algorithms for identifying tabular grids and merged cells within document layouts.
• Image-Based Table Extractors - Ships a pipeline that extracts table structures from images and outputs them as formatted HTML.
• Pipeline Configurations - Loads pipeline behavior from declarative YAML configuration files to override default settings for initialization and inference.
• Document Layout Extraction - Identifies and classifies document regions like titles, tables, and figures.
• Pipeline Execution Interfaces - Executes pre-built processing pipelines by specifying name, input file, and target device in a single terminal command.
• Pipeline REST API Servers - Starts a REST server from a pipeline name or config file, exposing models for inference over HTTP.
• Unified Pipeline Interfaces - Provides a unified Python, CLI, and REST API interface for executing all pre-built AI pipelines.
• AI Model Production Deployment - Serves trained pipelines as REST APIs, containerized services, or edge-device binaries for production inference.
• Modular Pipeline Architectures - Organizes vision and language tasks as composable pipeline modules that can be chained together.
• Model Inference Deployment - Deploys trained models as high-performance inference services, containerized endpoints, or on edge devices.
• AI Pipeline Service Deployments - Ships a framework for deploying AI pipelines as production-ready HTTP APIs.
• Multi-Device Inference Switchers - Enables switching inference between GPU, NPU, XPU, and MLU accelerators by modifying a single parameter across all pipelines.
• Image-Based Table Recognizers - Provides a pipeline that extracts structured table data from images and PDFs into multiple output formats.
• YAML-Driven Configurations - Controls pipeline behavior, model selection, and hardware acceleration through declarative YAML files.
• Human - Identifies and locates specific body joints like shoulders and elbows in images to analyze human pose.
• Instance Segmentation Engines - Detects each object instance and produces pixel-level masks for them.
• CLI Executions - Provides a CLI command and Python SDK for running instance segmentation on images.
• Instance Segmentation Service Deployments - Exposes the segmentation pipeline as an HTTP API returning detection boxes, masks, and scores.
• Multi-Device Inference Switching - Supports switching between GPU, NPU, XPU, and MLU accelerators with a single parameter.
• Rotated Detections - Detects objects with rotated bounding boxes that include angle information for reduced background noise.
• CLI Executions - Ships a CLI command that detects objects with rotated bounding boxes including angle information.
• Semantic Segmentation Training - Provides pre-trained models for pixel-level semantic segmentation of images into categories.
• Recognition Module Fine-Tuners - Provides a pipeline for retraining recognition modules on custom data to improve accuracy for domain-specific documents.
• High-Performance Pipeline Deployments - Optimizes model inference and pre/post-processing for faster end-to-end throughput in production.
• Face Detection - Locates human faces in images with bounding boxes using pre-trained models.
• Confidence Filtering - Applies global or per-class score thresholds to discard low-confidence bounding boxes from detection results.
• Pipeline Module Selectors - Ships a mechanism for selecting specific pre-trained models for individual sub-tasks within a pipeline.
• Feature Extractors - Computes compact vector representations of faces for recognition or verification tasks.
• GPU-Accelerated Inference - Supports GPU-accelerated inference via OpenCL to improve model prediction performance on supported hardware.
• CLI Executions - Offers a CLI command that runs image classification inference on images or directories.
• OCR Hardware Switching - Changes the compute device for the OCR pipeline by setting a single parameter with no code changes.
• Multi-Backend Inference Support - Supports switching between Paddle Inference, TensorRT, OpenVINO, ONNX Runtime, and Ascend OM.
• Inference Performance Optimization - Provides a high-performance inference plugin to accelerate model predictions and reduce production latency.
• Detection Pipeline Web Services - Exposes the object detection pipeline as an HTTP API that accepts images and returns structured detection results.
• Pipeline-Level Retrainers - Provides a pipeline for fine-tuning entire pre-trained pipelines on custom datasets to improve task-specific accuracy.
• Modular Fine-Tuning Pipelines - Allows retraining individual pipeline sub-modules on custom data and swapping them in.
• Fine-Tuned Pipeline Verifiers - Provides a pipeline for testing fine-tuned models by swapping them into the inference pipeline and verifying performance.
• Rotated Object Detection Fine-Tuners - Ships a pipeline for retraining rotated object detection models on custom data to improve accuracy for specific scenarios.
• Small Object Detection Fine-Tuners - Provides a pipeline for retraining small-object detection models on custom data to improve domain-specific accuracy.
• Vehicle Detection Fine-Tuners - Ships a pipeline for retraining vehicle detection and attribute recognition models on private data to improve accuracy.
• On-Device Deployments - Ships a detection pipeline that runs directly on Android and other edge devices without a remote server.
• Automatic Backend Selectors - Ships a high-performance inference plugin that automatically selects the optimal backend and configuration for model predictions.
• High-Throughput Inference Services - Wraps extraction pipelines as network-accessible services with optimized inference for production environments.
• OCR Model Fine-Tuners - Ships a pipeline for retraining OCR text detection and recognition modules on private datasets.
• Image Anomaly Detection Pipelines - Identifies defects or abnormal regions in images using pre-trained anomaly detection models.
• Pedestrian Attribute Pipeline Deployments - Packages the detection and attribute recognition pipeline as an HTTP API service for remote inference.
• Multi-Modal 3D Detections - Fuses camera and LiDAR data to detect and classify objects in 3D space for autonomous driving.
• Detection Service Deployments - Exposes rotated object detection as an HTTP API service.
• Small Object Detectors - Identifies and classifies small-sized objects in complex scenes for surveillance or autonomous driving.
• CLI Executions - Ships a CLI command that runs open-vocabulary segmentation using text or point prompts.
• Web APIs - Exposes object detection as an HTTP service that accepts images and prompts, returning detected objects and visualizations.
• CLI Executions - Ships a CLI command that runs open-vocabulary object detection using natural language prompts.
• Segmentation Service Deployments - Serves segmentation pipelines as HTTP services with prompt-based inputs.
• Time Series Anomaly Detection - Ships a pre-trained time series anomaly detection pipeline that identifies abnormal points in temporal data.
• Time Series Classification - Provides a time series classification pipeline that assigns category labels based on temporal patterns.
• Multi-Modal Detections - Implements a BEVFusion-based pipeline that fuses camera images with LiDAR point clouds for 3D object detection.
• CLI Executions - Ships a CLI command that runs 3D multi-modal object detection from fused camera and LiDAR data.
• Document Question Answering - Answers natural-language questions by reading and reasoning over the content of document images.
• Prompt-Based Segmentations - Implements prompt-based segmentation using text, boxes, or points to isolate objects without predefined categories.
• Document Analysis Fine-Tuners - Ships a pipeline for retraining document extraction models on custom data to improve accuracy for specialized layouts.
• Document Extraction Fine-Tuners - Provides a pipeline for retraining document extraction models on custom data to improve accuracy for specific document types.
• Instance Segmentation Fine-Tuners - Provides a pipeline for retraining instance segmentation models on custom data to improve accuracy for specific scenes.
• Semantic Segmentation Fine-Tuners - Ships a pipeline for retraining semantic segmentation models on custom data to improve accuracy for specific scenes.
• Private Data Fine-Tuners - Ships a pipeline for training selected pipeline modules on custom private datasets to improve task-specific accuracy.
• Seal Recognition Fine-Tuners - Provides a pipeline for retraining seal recognition modules on private data to improve accuracy for specific scenarios.
• Table Recognition Fine-Tuners - Provides a pipeline for retraining table recognition modules on private data to improve accuracy for specific document types.
• Time Series Forecasting Fine-Tuners - Ships a pipeline for retraining time series forecasting models on custom data to improve prediction accuracy.
• Multi-Label Classifiers - Assigns multiple category labels to a single image by analyzing its content.
• Inference Backend Tuners - Provides per-backend parameter tuning for inference engines like TensorRT and OpenVINO to optimize runtime performance.
• Seal Recognition Pipelines - Provides a CLI command that detects and extracts curved text from seal images.
• Model Executions - Downloads a pre-compiled inference library and optimized model, then builds and executes a prediction binary on Android via ADB.
• Classification Pipeline Deployers - Packages the image classification pipeline as a web service or edge device binary for production deployment.
• Edge-to-Cloud Deployments - Packages pipelines for deployment on Android, Docker, and high-performance servers.
• Multi-Format Segmentation Deployments - Packages segmentation models as services, APIs, or edge binaries for production.
• Document Analysis Services - Wraps the full document extraction pipeline into a REST API for remote clients to submit images or PDFs and receive structured results.
• Classification Service Deployers - Packages the image classification pipeline as an HTTP service that accepts image URLs or base64 data and returns predictions.
• Forecasting Service Deployments - Exposes a time series forecasting pipeline as an HTTP API endpoint for remote inference.
• Document Recognition Service Deployments - Provides REST API deployment for formula recognition pipelines.
• Inference - Runs a production-grade serving SDK inside Docker containers with GPU and CPU support.
• Web Service Deployments - Exposes the detection pipeline as an HTTP API that accepts images and returns structured results.
• Pedestrian Attribute Detections - Locates pedestrians and identifies attributes like gender, age, and clothing in images.
• Pedestrian Detections - Locates people in images with bounding boxes optimized for human detection.
• Image Feature Extraction - Computes compact vector representations of images for similarity search and retrieval.
• Human Pose Detections - Detects key body joints in images to reconstruct human poses using pre-trained models.
• Formula Recognition Engines - Converts mathematical formula images into LaTeX source code.
• CLI Executions - Provides a CLI command that converts mathematical formula images into LaTeX code.
• Vehicle Detection - Locates vehicles in images with bounding boxes optimized for vehicle detection.
• Vehicle Attribute Recognition - Ships a pipeline that locates vehicles and recognizes their type, color, and license plate attributes.
• Attribute Recognition Service Deployments - Exposes vehicle attribute recognition as a network service via REST API.
• OCR Web Services - Exposes the OCR pipeline behind a REST API endpoint for remote clients to submit images and receive text results.

Star history