PaddleX

3.4.3 · active · verified Sat Apr 11

PaddleX is a low-code development tool built on the PaddlePaddle deep learning framework. It integrates numerous ready-to-use pre-trained models, facilitating full-process AI development from model training to inference. Supporting various mainstream hardware, PaddleX aims to assist AI developers in industrial practice. The current version is 3.4.3, with an active release cadence.

Warnings

breaking PaddleX 3.0.x requires PaddlePaddle version 3.0.0 or higher. Older PaddlePaddle versions are incompatible and will lead to runtime errors.
Fix: Ensure `paddlepaddle` or `paddlepaddle-gpu` is installed with version `3.0.0` or higher before installing PaddleX.
gotcha On Python 3.12 and newer, `setuptools` is no longer included by default in virtual environments. This can cause installation failures for PaddlePaddle and subsequently PaddleX.
Fix: Explicitly install `setuptools` (e.g., `pip install setuptools`) in your virtual environment before installing PaddlePaddle or PaddleX.
gotcha When using the high-performance inference plugin, ensure the installed TensorRT version is compatible with your PaddlePaddle's CUDA version. For example, PaddlePaddle with CUDA 11.8 requires TensorRT 8.x (specifically recommended 8.6.1.6).
Fix: Refer to the PaddleX documentation for compatible TensorRT versions corresponding to your PaddlePaddle/CUDA setup and install it manually if necessary.
deprecated In PaddleX 3.0, static graph models for inference now use `.json` files instead of the older `.pdmodel` format. Using the old format with newer PaddleX versions for static graph inference may not work as expected.
Fix: Ensure your models are updated to the `.json` format for static graph inference with PaddleX 3.0+. Convert older `.pdmodel` files if you encounter issues.

Install

pip install paddlex Basic Installation
pip install "paddlex[base]" Installation with Basic Features
pip install "paddlex[ocr]" Installation with OCR Features

Imports

create_pipeline
```
from paddlex import create_pipeline
```

Quickstart

This quickstart demonstrates how to initialize an OCR pipeline and perform inference on an image URL using the `create_pipeline` function and its `predict` method. The example uses a public image URL, but local file paths are also supported.

import os
from paddlex import create_pipeline

# Example for OCR pipeline
# Note: This quickstart assumes PaddlePaddle and necessary OCR dependencies are installed.
# You can replace 'OCR' with other supported pipeline names (e.g., 'GeneralInstanceSegmentation', 'ImageMultiLabelClassification').
# The 'input' can be a local path, directory, or URL to an image.
# For a real-world scenario, you might download an image or use a local file.
image_url = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png"

try:
    # Create an OCR pipeline instance
    pipeline = create_pipeline(pipeline="OCR")
    print("OCR pipeline created successfully.")

    # Predict using the pipeline
    # The 'predict' method returns a dictionary of results.
    output = pipeline.predict(image_url)

    # Print the prediction results
    print("Prediction successful. Output:")
    print(output)

except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure PaddlePaddle is installed and compatible, and relevant PaddleX dependencies (e.g., 'paddlex[ocr]') are installed.")

view raw JSON →