repository·main·Indexed 19 days ago
https://github.com/deepghs/imgutilsA specialized Python image processing library for anime-style imagery. It provides high-level tools for character, face, head, and person detection, as well as difference detection (Tachie), clustering via CCIP and lpips_clustering, and tagging using the wd14-v2 model. Additional features include lineart generation, monochrome and truncated file validation, RGBA segmentation with isnetis, and ASCII art rendering.
imgutils is a module designed to parse and manage the structure and versions of configuration files.The imgutils.metadata module provides utilities for extracting and managing metadata from images. It includes specialized functions for retrieving general image information, as well as specific metadata formats like EXIF, IPTC, and XMP.
Key sub-modules include:
geninfo: General image information extraction.lsb: Least Significant Bit (LSB) related metadata or steganography detection/extraction.imgutils.preprocess.torchvision module provides utilities to integrate torchvision transforms and parsing logic into the imgutils preprocessing pipeline. It allows for registering custom torchvision transforms and parsers, as well as creating and parsing transform configurations.imgutils.validate.safe module provides tools to perform safety checks on images. It includes functions to calculate a safety score and perform a boolean safety check, typically used to identify potentially unsafe or inappropriate content in images.The imgutils.preprocess module provides a collection of tools for image preprocessing. It is organized into several sub-modules based on the underlying backend or functional category:
torchvision library (typically for PyTorch-based workflows).transformers library.Use these sub-modules to prepare images for various computer vision tasks, such as model inference or data augmentation.
The imgutils.detect module provides a collection of specialized detection tools for identifying various elements within images. Depending on your use case, you can use specific submodules to detect:
You can convert anime images to line drawings using models from controlnet_aux.
canny: The fastest method, but provides average quality.lineart: Higher quality but consumes more computational resources.lineart_anime: Recommended as the most balanced choice for most use cases.imgutils supports object detection specifically optimized for anime-style content:
Install the stable version of imgutils from PyPI using pip. This package requires Python version 3.8 or higher.
pip install dghs-imgutilsYou can install imgutils using pip from PyPI.
For standard CPU-based processing, use the default installation. If your environment has an available GPU, install the [gpu] extra to enable higher performance models.
# Standard installation
pip install dghs-imgutils
# GPU-accelerated installation
pip install dghs-imgutils[gpu]To install the latest development version directly from the GitHub repository, use the following command:
pip install -U git+https://github.com/deepghs/imgutils@main