Source code for imgutils.edge.lineart

"""
Overview:
    Get edge with lineart model.

    Having the **best effect**, closest to the drawing lines,
    but consuming a large amount of memory and computing power at runtime.
"""
from functools import partial
from typing import Optional

import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download

from ._base import resize_image, cv2_resize, _get_image_edge
from ..data import ImageTyping, load_image
from ..utils import open_onnx_model, ts_lru_cache


def _preprocess(input_image: Image.Image, detect_resolution: int = 512):
    input_image = np.array(input_image, dtype=np.uint8)
    input_image = resize_image(input_image, detect_resolution)
    return (input_image / 255.0).transpose(2, 0, 1)[None, ...].astype(np.float32)


@ts_lru_cache()
def _open_la_model(coarse: bool):
    return open_onnx_model(hf_hub_download(
        'deepghs/imgutils-models',
        f'lineart/{"lineart.onnx" if not coarse else "lineart_coarse.onnx"}',
    ))


[docs]def get_edge_by_lineart(image: ImageTyping, coarse: bool = False, detect_resolution: int = 512): """ Overview: Get edge mask with lineart model. :param image: Original image (assuming its size is ``HxW``). :param coarse: Use coarse model. In the coarse model, the lines will be deeper and richer, but the probability of extra lines or content appearing will increase. :param detect_resolution: Resolution when passing the image into neural network. Default is ``512``. :return: A mask with format ``float32[H, W]``. """ image = load_image(image, mode='RGB') output_, = _open_la_model(coarse).run(['output'], {'input': _preprocess(image, detect_resolution)}) output_ = cv2_resize(output_[0].transpose(1, 2, 0), image.width, image.height) return 1.0 - output_.clip(0.0, 1.0)
[docs]def edge_image_with_lineart(image: ImageTyping, coarse: bool = False, detect_resolution: int = 512, backcolor: str = 'white', forecolor: Optional[str] = None): """ Overview: Get an image with the extracted edge from ``image``. :param image: Original image (assuming its size is ``HxW``). :param coarse: Use coarse model. In the coarse model, the lines will be deeper and richer, but the probability of extra lines or content appearing will increase. :param detect_resolution: Resolution when passing the image into neural network. Default is ``512``. :param backcolor: Background color the target image. Default is ``white``. When ``transparent`` is given, \ the background will be transparent. :param forecolor: Fore color of the target image. Default is ``None`` which means use the color \ from the given ``image``. :return: An image with the extracted edge from ``image``. Examples:: .. image:: lineart.plot.py.svg :align: center When ``coarse`` is used: .. image:: lineart_coarse.plot.py.svg :align: center """ return _get_image_edge( image, partial(get_edge_by_lineart, coarse=coarse, detect_resolution=detect_resolution), backcolor, forecolor )