How to Set Up ModelScope with Anaconda and Run OCR Inference via PHP

What Is ModelScope?

ModelScope is an open‑source model‑as‑a‑service platform that provides a one‑stop solution for AI developers, offering flexible, easy‑to‑use, low‑cost model services for a wide range of tasks.

Environment Installation

System Requirements

Operating System: Linux

Environment manager: Anaconda

Python version: 3.10 Deep‑learning framework: PyTorch

Compute platform: CPU

Install Anaconda

Anaconda bundles over 190 scientific packages and a conda package manager, simplifying Python package installation and environment management.

wget https://repo.anaconda.com/archive/Anaconda3-2023.09-0-Linux-x86_64.sh

Run the installer and accept the license:

sh Anaconda3-2023.09-Linux-x86_64.sh
# accept the license terms --yes

Verify the installation:

conda --version

Create and Activate a Python Environment

conda create -n tinywan-modelscope python=3.10

conda activate tinywan-modelscope

Common conda commands:

# List environments
conda env list
# Install a package, e.g., numpy
conda install numpy
# Remove a package
conda remove package_name
# Deactivate environment
conda deactivate
# Delete an environment
conda remove -n your_env_name --all

Install PyTorch

pip3 install torch torchvision torchaudio

Install ModelScope Library

Install the core framework: pip3 install modelscope For domain‑specific capabilities, install extra dependencies. Example for multimodal models: pip install "modelscope[multi-modal]" Example for NLP models:

pip3 install "modelscope[nlp]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html

Verify Installation with an NLP Pipeline

Run a Chinese word‑segmentation task to confirm the setup:

python -c "from modelscope.pipelines import pipeline; print(pipeline('word-segmentation')('今天天气不错，适合 出去游玩。开源技术小栈'))"

Using PHPY to Call ModelScope (OCR Example)

PHPY enables PHP code to import and use Python packages, allowing PHP to run ModelScope pipelines.

PHP Code Example (captcha.php)

<?php
/**
 * @link https://modelscope.cn/models/damo/cv_convnextTiny_ocr-recognition-general_damo/summary
 */
$pipeline = PyCore::import('modelscope.pipelines')->pipeline;
$Tasks = PyCore::import('modelscope.utils.constant')->Tasks;
// Use the OCR model
$pipe = $pipeline($Tasks->ocr_recognition, model: 'damo/cv_convnextTiny_ocr-recognition-general_damo');
$file = '/tmp/captcha.png';
file_put_contents($file, file_get_contents('https://business.swoole.com/page/captcha_register'));
echo '识别结果：' . $pipe($file)['text'][0], PHP_EOL;
?>

The official example can be found at:

https://github.com/swoole/phpy/blob/main/examples/modelscope/captcha.php

Model Details

Task: Text recognition from images.

Architecture: Convolutional Backbone extracts visual features. ConvTransformer Blocks perform contextual modeling. CTC loss decodes the output and optimizes gradients.

Required Dependency: OpenCV

The OCR model relies on the OpenCV library for image processing.

pip install opencv-python

Run the PHP Script

/usr/local/php-8.2.14/bin/php captcha.php

Sample output:

识别结果：7R2N

Execution Log (excerpt)

(tinywan-modelscope) $ /usr/local/php-8.2.14/bin/php captcha.php
2024-01-19 22:41:20,505 - modelscope - INFO - PyTorch version 2.1.2 Found.
2024-01-19 22:41:20,506 - modelscope - INFO - Loading ast index from /home/www/.cache/modelscope/ast_indexer
... (log continues) ...
识别结果：电子元器件提供BOM配单

These steps provide a complete end‑to‑end setup for installing ModelScope, adding the OCR model, and invoking it from PHP via PHPY.