sketchkit.image2sketch.hed package

Submodules

sketchkit.image2sketch.hed.model module

HED (Holistically Nested Edge Detection) integration.

This module wraps the original Caffe implementation using OpenCV’s DNN loader. It mirrors the public API of other image2sketch backends so callers can request HED outputs via image2sketch(..., method="hed").

class sketchkit.image2sketch.hed.model.HEDModel(*, checkpoint_dir: str | PathLike | None = None, auto_download: bool = True, postprocess: bool = True, postprocess_threshold: float = 0.09803921568627451, postprocess_small_edge: int = 5)

Bases: object

High level wrapper around the original HED Caffe model.

MEAN_BGR = (104.00698793, 116.66876762, 122.67891434)

MODEL_FILENAME = 'hed_pretrained_bsds.caffemodel'

MODEL_SHA1 = '2c5d7842f25f880eec62fc610b500c5cf2aa351d'

MODEL_URL = 'https://vcl.ucsd.edu/hed/hed_pretrained_bsds.caffemodel'

static _download_weights(url: str, destination: Path, expected_sha1: str) → None

_infer_single(net: Any, image: Image, size: Tuple[int, int]) → Image

_initialise_net() → Any

_net: Any | None

static _to_pil_image(image: Image | ndarray | str | PathLike) → Image

static _validate_sha1(path: Path, expected_hex: str) → bool

property checkpoint_dir: Path

ensure_assets() → None

Ensure the pretrained weights are available locally.

generate(image: Image | ndarray | str | PathLike | Sequence[Image | ndarray | str | PathLike], *, size: int | Tuple[int, int] | None = None) → Image | List[Image]

Generate edge maps for the provided image(s).

sketchkit.image2sketch.hed.postprocess module

HED post-processing pipeline

class sketchkit.image2sketch.hed.postprocess.PostprocessParams(threshold: float = 0.09803921568627451, small_edge: int = 5, nms_radius: int = 1, nms_border: int = 5, nms_multiplier: float = 1.01, smooth_radius: float = 4.0)

Bases: object

Configuration mirroring the defaults from PostprocessHED.m.

nms_border: int = 5

nms_multiplier: float = 1.01

nms_radius: int = 1

small_edge: int = 5

smooth_radius: float = 4.0

threshold: float = 0.09803921568627451

sketchkit.image2sketch.hed.postprocess.apply_postprocess(edge_map: ndarray, *, params: PostprocessParams = PostprocessParams(threshold=0.09803921568627451, small_edge=5, nms_radius=1, nms_border=5, nms_multiplier=1.01, smooth_radius=4.0)) → ndarray

Apply the official HED post-processing steps.

Parameters:

• edge_map – Single-channel array in the range [0, 1] representing raw HED edge probabilities.

• params – Tunable parameters. Defaults replicate the MATLAB script.

Returns:

Float32 array in [0, 1] where 1 denotes white background and 0 denotes edge strokes.

Return type:

np.ndarray

Module contents

Public entry points for the HED image2sketch backend.

class sketchkit.image2sketch.hed.HEDModel(*, checkpoint_dir: str | PathLike | None = None, auto_download: bool = True, postprocess: bool = True, postprocess_threshold: float = 0.09803921568627451, postprocess_small_edge: int = 5)

Bases: object

High level wrapper around the original HED Caffe model.

MEAN_BGR = (104.00698793, 116.66876762, 122.67891434)

MODEL_FILENAME = 'hed_pretrained_bsds.caffemodel'

MODEL_SHA1 = '2c5d7842f25f880eec62fc610b500c5cf2aa351d'

MODEL_URL = 'https://vcl.ucsd.edu/hed/hed_pretrained_bsds.caffemodel'

static _download_weights(url: str, destination: Path, expected_sha1: str) → None

_infer_single(net: Any, image: Image, size: Tuple[int, int]) → Image

_initialise_net() → Any

_net: Any | None

static _to_pil_image(image: Image | ndarray | str | PathLike) → Image

static _validate_sha1(path: Path, expected_hex: str) → bool

property checkpoint_dir: Path

ensure_assets() → None

Ensure the pretrained weights are available locally.

generate(image: Image | ndarray | str | PathLike | Sequence[Image | ndarray | str | PathLike], *, size: int | Tuple[int, int] | None = None) → Image | List[Image]

Generate edge maps for the provided image(s).