ifc2plan

Turn an IFC building model into 2D floor plans — as images you can look at, or as data you can analyse.

For each storey of a model, ifc2plan cuts a horizontal section through the building and writes:

• a floor plan image (PNG), in one of four drawing styles, optionally coloured by room type
• a table of the shapes (CSV), one row per element, so you can measure, count, or feed it into other tools

The ground floor of the example building included in this repository — produced by step 3 below, unedited.

Install

You need Python 3.9 or newer.

git clone https://github.com/datarefinerylab/ifc2plan.git
cd ifc2plan
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

On Windows, use .venv\Scripts\activate instead of source .venv/bin/activate.

Run it

Stay in the ifc2plan folder and give the full path to the script, exactly as shown below. An example building is included, so you can copy these lines as they are.

1. Look inside the file first. This is instant — it lists the storeys and how many elements each one has, and tells you the storey numbers you'll need.

python src/ifc2plan/extract_floor_plans.py "examples/data/Shependomlaan/IFC Schependomlaan.ifc" --overview

2. Draw one floor plan. Storey 1 is the ground floor of this building — step 1 told you that.

python src/ifc2plan/extract_floor_plans.py "examples/data/Shependomlaan/IFC Schependomlaan.ifc" \
  --storey 1 --formatter image

3. Draw it with rooms coloured by type, and get the data too. This is the drawing at the top of this page.

python src/ifc2plan/extract_floor_plans.py "examples/data/Shependomlaan/IFC Schependomlaan.ifc" \
  --storey 1 --formatter image wkt --colored-spaces \
  --naming-conversion naming_conversion.csv

Results appear in output/IFC Schependomlaan/.

To use your own model, put its path in place of the example. Start with --overview, then one storey — a whole building takes a while, and there is no point waiting for it until you know the settings are right. Leave --storey off to do every storey.

--storey takes more than one number, so you rarely have to run the command twice:

You type	You get
--storey 1	just that storey
--storey 0,2	those two
--storey 1-3	storeys 1, 2 and 3
--storey begane	every storey whose name contains "begane" — copy it off the --overview list
--storey all	the whole building (same as leaving it off)

If you mistype it, the command stops and prints the storeys the model actually has, so you can fix the line without going back to --overview.

On the long command: always type the full src/ifc2plan/extract_floor_plans.py. The short python extract_floor_plans.py fails unless you have moved into that folder first. If you get can't open file or ModuleNotFoundError, this is why.

Options

Option	What it does	Default
--overview	List storeys and element counts, then stop. Costs nothing — always start here	off
--storey	Which storeys to draw — a number from --overview, a list (0,2), a range (1-3), part of a storey's name, or all	all storeys
--formatter	What to write: image, wkt, or both (--formatter image wkt)	wkt
--output	Folder to write into	output
--style	Look of the drawing: professional, minimal, colorful, technical	professional
--colored-spaces	Give each room type its own colour (needs --naming-conversion)	off
--both	Write a coloured and a black & white version	off
--space-only	Draw only the rooms, leaving out walls, doors, and everything else	off
--naming-conversion	CSV that translates room names — see Room names	none
--width, --height	Image size in pixels	2048
--section-offset	How high above each floor to cut, in metres	1.5
--parallel	Use several processor cores at once	off
--max-faces	Skip unusually detailed elements to save time — see If a run is slow	none
--max-elements	Stop after this many elements. Useful for a quick trial run on a big file	none
--slow-element-seconds	Report any element that takes longer than this to convert	5.0
--skip-failed	Keep going when an element fails instead of stopping	off

What you get

Files are named after the storey as the model names it, inside a folder named after the model. Running example 3 above gives you:

output/
└── IFC Schependomlaan/
    ├── 00 begane grond_floor_plan.png   # the drawing
    └── 00 begane grond_floor_plan.csv   # the data

With --both you get two drawings per storey instead of one, marked _colored and _bw:

    ├── 00 begane grond_floor_plan_colored.png
    └── 00 begane grond_floor_plan_bw.png

Each row of the CSV is one element, with these columns:

Column	Meaning
type	What it is in the model — IfcWall, IfcSpace, IfcDoor, …
name	The element's name from the model
room_type	The tidied, translated room type
room_type_original	The room name exactly as the model spells it, so rooms you haven't translated are still identifiable
geometry	The outline itself, as WKT — a text format that QGIS, PostGIS, Shapely, and R can all read directly

Room names

Models often name rooms in the local language. Give ifc2plan a two-column CSV and it will translate them, which is also what lets it colour rooms by type. naming_conversion.csv in this repository is a Dutch → English starting point:

original,english
badkamer,bathroom
slaapkamer,bedroom
woonkamer,livingroom
keuken,kitchen
gang,corridor
berging,storage
balkon,balcony

Add your own rows in the same format. Matching ignores capitals. Rooms with no matching row still appear in the output — you'll see them under their original name in the room_type_original column.

Drawing styles

--style	Looks like
professional (default)	Clean architectural drawing with subtle fills
minimal	Black and white; rooms separated by shades of grey
colorful	Bright and saturated, for slides and posters
technical	Outlines only, no fills, like a drafting sheet. Room colouring does not apply

Every style can be combined with --colored-spaces except technical.

📸 Compare the styles

All three are the ground floor of the bundled model, so you can reproduce any of them by swapping the --style and --colored-spaces settings.

Professional — the default. Rooms share one fill; doors and windows are picked out in colour.

Professional with --colored-spaces — each room type gets its own colour and a legend entry.

Technical — outlines only, like a drafting sheet.

Troubleshooting

can't open file 'extract_floor_plans.py' or ModuleNotFoundError Use the full path: python src/ifc2plan/extract_floor_plans.py, run from the ifc2plan folder.

No images came out Add --formatter image. On its own, the tool writes CSV only.

No floor plans at all The model needs IfcBuildingStorey elements — --overview will show you whether it has any.

Rooms aren't coloured You need --colored-spaces and --naming-conversion, and the room names in the model have to match rows in your CSV.

It ran out of memory A model takes roughly six times its file size in memory, so a 200 MB file wants about 1.2 GB. Do one storey at a time, or use --max-elements to cap the work. --parallel is faster but holds one copy of the model per core, so it needs more memory, not less.

It's taking forever — see below.

If a run is slow

Almost always, a handful of elements are responsible. On one real storey, 4 elements out of 669 took 99% of the time — over half an hour between them.

ifc2plan names them as it goes and lists them at the end of each storey:

   ⏱  4 slow element(s), 309s total:
        157.8s  IfcCovering #462475 '47_GM_waterslag' (26,011 faces)
         86.5s  IfcCovering #504259 '47_GM_waterslag' (26,011 faces)

If you don't need those elements, --max-faces skips them. On the storey above, --max-faces 12000 cut the drawing step from 162 seconds to 5.6 and left out exactly those four elements.

Read the report before choosing a number. Face count only roughly predicts cost: in the same model a 10,239-face door converted in 0.66 s while a 15,736-face covering took 23 s. Set the limit too low and you lose geometry without saving time. This is why the option is off by default — it changes what ends up in your drawing.

Which elements does --max-faces actually count?

Face counts are read from the model's own representation data, covering tessellated bodies (IfcPolygonalFaceSet, IfcTriangulatedFaceSet), brep solids (IfcFacetedBrep and the rest of the IfcManifoldSolidBrep family), surface models, and mapped items wrapping any of those.

Anything else counts as zero and is therefore never skipped — an unfamiliar body type makes the flag do less, never more. Swept solids such as IfcExtrudedAreaSolid declare no face count at all, so they are always converted.

Origin

ifc2plan began as a fork of byildiz/BatchPlan (MIT licensed) and was renamed after diverging substantially. It is a derivative work and the original copyright is retained — see LICENSE.

What changed since the fork

• Derived from byildiz/BatchPlan@4958a6f. This repository's history was truncated to its own work, so the original commits are not in this history — they remain in byildiz/BatchPlan. The root commit here records the derivation.
• The OpenCASCADE/SWIG geometry pipeline was replaced with a Trimesh
  • Shapely engine.
• The original's material/LCA database tooling was removed; this repository is scoped to floor plan and geometry extraction.
• The former repository, datarefinerylab/BatchPlan, is archived and read-only.

---

Built with: IfcOpenShell • Trimesh • Shapely