procslib 0.5.0

Multi-purpose processing library for downstream use

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:

Key	Description
twitter_logfav	Predicts log-scaled Twitter favorites for anime images.
weakm_v2	Aesthetic prediction model for anime images using WeakM v2 scoring.
weakm_v3	Updated WeakM v3-based aesthetic scoring model for anime images.
siglip_aesthetic	A Siglip-based model for aesthetic prediction (requires specific transformers versions).
pixiv_compound_score	Predicts a compound aesthetic score for Pixiv-based anime images.
cv2_metrics	Computes basic image quality metrics (noise, brightness, contrast, sharpness, etc.).
complexity_ic9600	Predicts image complexity using the IC9600 model.
rtmpose	Detects human pose keypoints in images using RTMPose.
depth	Uses MiDaS-based depth estimation to provide a "depthness" score (0.0-1.0).
q_align_quality	Predicts image quality scores using the QAlign model.
q_align_aesthetics	Predicts image aesthetics using the QAlign model.
laion_watermark	Detects watermarks in images using a model from LAION.
clip_aesthetic	Uses CLIP-based embeddings for aesthetic scoring and zero-shot classification.
vila	Generates textual descriptions of images using the NVILA-15B model.
jz_tagger	Multi-label image classification model with aesthetic scoring (Danbooru-based).
aigc_classifier	Classifies images as AI-generated or real using incantor/aigc_real_cls.
szh_image_category	Categorizes images using szh/image_category_cls.
anime_real_cls	Classifies images as anime or real with confidence scores using incantor/anime_real_cls.

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