vectorify 0.1.0

Convert raster images (PNG, JPG, BMP, WebP) to simplified SVG vector graphics using k-means colour quantization and contour tracing.

ive am

Convert raster images (PNG, JPG, etc.) into simplified SVG vector graphics using k-means colour quantization and contour tracing.

Installation

Requires Python 3.11+.

uv sync

Usage

uv run vectorify generate <filepath> <output> [options]

Example

uv run vectorify generate photo.png ./output.svg --n_colors 8 --smooth 2 --simplify 0.5

Update README images

Regenerate all example SVGs, PNGs, and grid images used in this README:

uv run vectorify update-readme

Arguments

Argument	Type	Description
filepath	string	Path to the input image
out_dir	string	Directory for the output SVG

Options

Option	Type	Default	Description
--n_colors	int	8	Number of colours to quantize the image into
--median_size	float	5	Median filter kernel size for label smoothing (0 = disabled)
--min_region	float	50	Minimum connected region size in pixels (0 = disabled)
--smooth	float	2	Gaussian smoothing sigma applied to contours (0 = disabled)
--simplify	float	0.5	RDP simplification epsilon in pixels (0 = disabled)
--buffer	float	1	Outward normal offset to expand shapes and fill gaps

Configuration

buffer

Expands each shape outward along its edge normals by the given pixel amount. Eliminates sub-pixel gaps between adjacent colour regions caused by smoothing or simplification.

---

median_size

Applies a median filter of this kernel size to the label map. Fills small holes, removes straggler pixels, and smooths region boundaries. Larger values produce more aggressive cleanup.

---

min_region

Removes connected components smaller than this many pixels, replacing them with their most common neighbouring colour. Catches isolated specks that the median filter misses.

---

n_colors

Number of distinct colours the image is quantized into via k-means clustering. Lower values produce a more simplified, poster-like result.

---

simplify

Epsilon value for the Ramer-Douglas-Peucker algorithm. Reduces the number of vertices in each contour path. Higher values = fewer vertices = simpler shapes.

---

smooth

Gaussian smoothing sigma applied to contour vertices. Rounds off jagged pixel-aligned edges into flowing curves. Applied before simplification.