Sketchy

Sketchy is a SketchDataset loader for the SVG subset of The Sketchy Database. The dataset contains vector-based sketches categorized by class name. It natively handles pruning invalid or ambiguous sketches based on the dataset’s provided annotation text files.

Source: datasets/sketchy.py

Data Format

Each sample is an SVG file containing vector paths. The loader parses the XML tree and handles:

  • <path> elements: The d attribute is parsed to extract control points. Supported commands include Move and CubicBezier. Line segments are automatically converted into cubic Bézier representations.

  • Dimensions: Canvas dimensions are extracted from the SVG viewBox or fallback width/height attributes.

  • Unsupported elements: Primitive SVG shapes (like rect, circle, ellipse, line, polyline, polygon) and path transforms are ignored or unsupported by this parser.

Each path is converted into a Path composed of cubic Bézier Curves to remain consistent with SketchKit’s unified format.

Directory Layout

After download and extraction, the dataset is expected under:

<root>/
  sketches/
    airplane/
      n02691156_58-1.svg
      ...
      invalid.txt
    apple/
      ...
  sketches-06-04.7z
  .metadata.parquet

  • <root>/sketches/ contains subfolders for each class name.

  • Each class folder contains its respective .svg files and an invalid.txt file listing unidentifiable or erroneous sketches (which are tracked via the pruned metadata column).

Code

from sketchkit.datasets import Sketchy

ds = Sketchy(cislab_source=True)

sketch = ds[0]
print(sketch.width, sketch.height)
print(sketch.path_num, sketch.curve_num)

# Filter out pruned (invalid) sketches for a specific class
valid_airplanes = ds.items_metadata[
    (ds.items_metadata["class_name"] == "airplane")
    & (ds.items_metadata["pruned"] == False)
]
sketches = [ds[row.id] for _, row in valid_airplanes[:5].iterrows()]

Arguments

  • cislab_source: Selects the download source. If True, the dataset is downloaded from the CISLAB CDN mirror; otherwise it is downloaded from official host.

    • CISLAB mirror: https://cislab.hkust-gz.edu.cn/projects/sketchkit/datasets/Sketchy/sketches-06-04.7z

    • Official host: https://drive.google.com/file/d/1Qr8HhjRuGqgDONHigGszyHG_awCstivo

Metadata Columns

  • id: Global unique identifier across all sketches.

  • pruned: Boolean flag indicating if the sketch was listed in the category’s invalid.txt file.

  • category: The class category of the sketch (e.g., “airplane”).

  • filename: The specific SVG file name (e.g., “n02691156_58-1.svg”).

  • path: The absolute path to the SVG file.

  • sketch_name: The relative identifier combining class and filename (e.g., “airplane/n02691156_58-1.svg”).

  • split: Split in the dataset (e.g., “train”, “valid”, or “test”).