python-cgm

Read-only Python tools for parsing binary and clear-text CGM (ISO/IEC 8632) files, producing final SVG output with optional raster tile backgrounds, and extracting hotspot metadata as JSON.

This package focuses on practical CGM extraction workflows: parsing CGM content, extracting image-bearing Cell Array payloads, decoding clear-text tile arrays, composing raster+vector SVG output, and recovering hotspots from APD region properties and APS geometry fallback. It does not support writing CGM files.

Installation

Install the latest version using pip:

pip install python-cgm

What It Does

• Parses binary and clear-text CGM command streams.
• Finds Cell Array elements (class 4, element 9) and extracts their raw payload bytes.
• Decodes clear-text tiled bitonal, indexed, and direct-color arrays.
• Decodes prefixed class 4, id 29 raster payloads using a fixed sample-derived decode model.
• Builds a final SVG output that can include an embedded raster background.
• Converts vector-like CGM drawing primitives into SVG overlays.
• Extracts hotspots from APD name/region records and APS geometry groups.
• Exports parsed element data, payload metadata, rendered SVG, and hotspots as JSON.

Quick Start

from cgm import (
    extract_data_json,
    extract_final_image_and_hotspots,
    extract_hotspots,
    extract_raw_images,
    extract_raw_images_to_directory,
    extract_vector_svg,
)

images = extract_raw_images("drawing.cgm")
print(f"Found {len(images)} raster payload(s)")

for image in images:
    print(
        image.index,
        image.element_offset,
        image.width,
        image.height,
        len(image.payload),
    )

written = extract_raw_images_to_directory("drawing.cgm", "./out")
print("Wrote", len(written), "payload file(s)")

svg = extract_vector_svg("drawing.cgm")
print("SVG length:", len(svg))

snapshot_json = extract_data_json("drawing.cgm")
print("JSON length:", len(snapshot_json))

final = extract_final_image_and_hotspots("drawing.cgm")
print("Final SVG length:", len(final["image"]))
print("Hotspots:", len(final["hotspots"]))

hotspots = extract_hotspots("drawing.cgm")
print("Hotspot objects:", len(hotspots))

CLI

After installation, use the CLI to export the final SVG and hotspot JSON:

cgm-extract file.cgm ./out

By default this writes:

• <basename>_0000.svg
• <basename>_0000.hotspots.json

With debug enabled it also writes:

• <basename>_decode_report.json

Optional flag:

cgm-extract file.cgm ./out --debug

API

• extract_raw_images(file_path) -> list[RawImage]
• extract_raw_images_from_bytes(data) -> list[RawImage]
• extract_raw_images_to_directory(file_path, output_dir, stem="image") -> list[Path]
• extract_rendered_images_to_directory(file_path, output_dir, stem="image", debug_report=False) -> list[Path]
• extract_vector_svg(file_path) -> str
• extract_vector_svg_from_bytes(data) -> str
• extract_vector_svg_to_directory(file_path, output_dir, stem="image") -> Path
• extract_data_json(file_path) -> str
• extract_data_json_from_bytes(data) -> str
• extract_data_json_to_directory(file_path, output_dir, stem="image") -> Path
• extract_hotspots(file_path) -> list[HotSpot]
• extract_hotspots_from_bytes(data) -> list[HotSpot]
• extract_hotspots_to_directory(file_path, output_dir, stem="image") -> Path
• extract_final_image_and_hotspots(file_path) -> dict[str, object]

RawImage fields:

• index: zero-based image index.
• element_offset: byte offset of the CGM element in the source file.
• payload: raw image payload bytes.
• width / height: dimensions when present in common binary or clear-text tile layouts.
• local_color_precision: declared color precision for the payload when available.
• cell_representation_mode: declared cell representation mode when available.

Supported CGM Elements And Features

The module focuses on practical extraction/rendering coverage for common binary and clear-text CGM workflows.

Binary CGM Element Coverage

• class 1, id 3 (VDC Type): used to choose coordinate decoding path.
• class 1, id 10 (Color Value Extent): used for 16-bit direct-color scaling.
• class 1, id 11 (VDC Integer Precision): used for strict integer VDC decode.
• class 1, id 12 (VDC Real Precision): used for strict real VDC decode.
• class 2, id 6 (VDC Extent): used to set SVG view extents and raster placement.
• class 3, id 4 (Transparency): mapped to SVG background behavior.
• class 3, id 5 (Clip Rectangle): mapped to SVG clip paths.
• class 3, id 6 (Clip Indicator): enables/disables clipping.
• class 4, id 1 (Polyline): rendered to SVG polylines.
• class 4, id 2 (Disjoint Polyline): rendered as segment polylines.
• class 4, id 3 (Polymarker): rendered as SVG marker circles.
• class 4, id 4 (Text continuation context): appended to prior text where applicable.
• class 4, id 5 (Text): rendered to SVG text.
• class 4, id 6 (Append Text): appended to prior text runs.
• class 4, id 7 (Polygon): rendered to SVG polygons.
• class 4, id 8 (Polygon Set): rendered as polygon geometry.
• class 4, id 9 (Cell Array): extracted as RawImage payloads and used as raster candidates.
• class 4, id 10 and id 26 (GDP-like primitives): decoded as polyline-style vectors.
• class 4, id 11 (Rectangle): rendered as SVG rect.
• class 4, id 12 (Circle): rendered as SVG circle.
• class 4, ids 13-16, 18-25, 27 (arc families): rendered as best-effort polyline geometry.
• class 4, id 17 (Ellipse): rendered as SVG ellipse.
• class 4, id 28: parsed but no strict vector fallback rendering.
• class 4, id 29 (Restricted Text or modeled prefixed raster payload): restricted text is rendered when text payload decodes; non-text payloads are raster-decoded only when their prefix matches known sample-derived profiles.
• class 5, id 3 (Line Width): applied to SVG stroke width.
• class 5, id 4 (Line Color): applied via palette/index mapping.
• class 5, id 15 (Character Height): applied to SVG text size.
• class 5, id 34 (Color Table): used for indexed palette and color mapping.
• class 9, id 1 (Application Data / APD): used for hotspot name/region extraction.
• class 0, id 21/22/23 (APS begin/end forms): used for hotspot grouping.

Clear-Text Command Coverage

• Vector primitives: LINE, POLYLINE, DISJOINTPOLYLINE, POLYMARKER, POLYGON, POLYGONSET, RECTANGLE, CIRCLE, ARC3PT, ARCCENTRE, ELLIPSE, ELLIPARC, GDP.
• Text primitives: TEXT, APPENDTEXT, RESTRICTEDTEXT.
• Raster/tile commands: CELLARRAY, BEGTILEARRAY/ENDTILEARRAY, BITONALTILE, MONOCHROMETILE, INDEXCOLORTILE, COLORTILE, DIRECTCOLORTILE (and colour spelling variants).
• Attributes/control: VDCEXT, COLRVALUEEXT, COLRTABLE, LINECOLR, TRANSPARENCY, CLIPRECT, CLIPIND.
• Hotspot-related data: BEGAPS, APD, ENDAPS.

Raster Decoding Features

• Extracts raw Cell Array payload bytes with metadata where present.
• Decodes bitonal raster data for uncompressed, CCITT Group 3, and CCITT Group 4 paths.
• Decodes indexed-color and direct-color tile payloads when dimensions/precision are usable.
• Composes raster backgrounds into SVG (embedded PNG data URI) before vector overlays.
• For multi-payload class 4, id 29 rasters, emits separate SVG <image> tile overlays when a simple inferred tile grid decodes successfully.

Hotspot Features

• Extracts APD name and region records into hotspot JSON.
• Falls back to APS geometry-based bounding boxes when explicit region data is absent.

Scope And Limitations

• The exact supported CGM elements/commands are listed in the Supported CGM Elements And Features section above.
• This project is extraction-oriented: it parses and exports data/SVG/JSON, but does not support CGM authoring or round-trip editing.
• Rendering is best-effort for many real-world files; unsupported or profile-specific constructs may be skipped rather than guessed from heuristic rewrites.
• Raster composition is metadata-dependent. Clear-text tile arrays are composed directly; binary Cell Array payloads are extracted and used as raster candidates.
• Raster decoding requires runtime dependencies Pillow and imagecodecs (installed with this package by default).
• JSON exports can be large because they include full element parameter and payload hex data.

License

python-cgm (C) 2026 Kestin Goforth.

This project is licensed under the BSD 3-Clause License - see the license file for details.