procslib

Multi-purpose processing library for various inference tasks.
Generated with copier-uv.

Installation

pip install procslib

Or using uv:

uv tool install procslib

Note: The inference requirements are not included for faster unittesting. See Dev Guide for a proper inference setup.

Quick Usage

Below is a minimal example of how to infer images with procslib:

from procslib import get_model_keys, get_model

# List available models
print(get_model_keys())

# Create a model, e.g. "twitter_logfav"
model = get_model("twitter_logfav")

# Infer on some images
image_paths = ["path/to/image1.jpg", "path/to/image2.jpg"]
results_df = model.infer_many(image_paths)
print(results_df.head())

Supported Models

You can retrieve a model via get_model(key). Here’s a quick reference:

Note: Q-Align and Siglip Aesthetics are incompatible with each other’s transformers version. If you need both, see Docs: Handling Conflicting Dependencies.

Development

For development tasks (testing, formatting, releasing), see Dev Guide or run:

make setup   # one-time
make format  # auto-format
make test
make check
make changelog
make release version=x.y.z

To build wheels manually, run the following commands:

python -m pip install build twine
python -m build
twine check dist/*
twine upload dist/*

Documentation

To learn more, visit our MkDocs-based docs or run:

make docs host=0.0.0.0

• Quick Start: Using Models
• Quick Start: Adding New Models
• VSCode Tips
• Detailed metric descriptions