Skip to content
Other

FloorPlanCAD YOLO11n

Architectural Floor Plan Element Recognition

2 min readupdated

A collection of fine-tunes based on the compact YOLO11n architecture (Ultralytics), trained for automated analysis of 2D blueprints and floor plans from the FloorPlanCAD dataset. The release includes two core models: an object detector for spotting 30 classes of elements (doors, windows, furniture, plumbing fixtures, stairs) and a segmentation model for isolating walls, curtain walls, and railings. Unlike general-purpose detectors, these weights are tailored specifically for black-and-white line drawings and can serve as the first step in converting 2D floor plans into 3D scenes.

Key features

  • Two specialized models: runs/d11n_s1/best.pt handles the detection of 30 types of interior and structural elements, while runs/w11n_s1/best.pt performs pixel-level segmentation of walls and railings.
  • Built for line drawings: The models take black-and-white images (black lines on a white background across 3 identical channels) scaled at roughly 10 mm per pixel, and support direct conversion from raw RGBA PNGs in FloorPlanCAD.
  • Fair benchmark against YOLOv8n: Tested across three seeds (seed 0, 1, 2) at 1024 resolution, showing that YOLO11n and the baseline YOLOv8n deliver statistically identical detection accuracy.

What you can do

  • Automate floor plan digitization: Combine the detector and segmentation model in 2D-to-3D pipelines to extract wall geometry and place 3D furniture and door assets across a floor plan.
  • Integrate into CAD tools or web apps: Run inference via the fpyolo.predict module on CPU or GPU for fast element counting and blueprint verification.

Example

python -m fpyolo.predict --d-weights runs/d11n_s1/best.pt --w-weights runs/w11n_s1/best.pt --w-conf 0.25 \
    --schema configs/classes_v0.yaml --input plan.png --out ./out --device 0
Read moreDetails
Detection (YOLO11n and YOLOv8n): core-class mAP50-95 reaches 0.725 ± 0.003 (the Δ +0.0004 difference is statistically insignificant).
Wall segmentation (YOLO11n-seg): wall pixel IoU hits 0.703 / 0.701 / 0.691 across three seeds.
Training setup: imgsz 1024 resolution, SGD optimizer (lr0 0.01), batch 32, up to 300 epochs (patience 50).
Limitations: Requires compliance with the FloorPlanCAD dataset terms of use, and inputs must strictly consist of black lines on a white background.
Discuss Telegram 0 113

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # linhjr87/2D-to-3D-yolov11n-models
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download linhjr87/2D-to-3D-yolov11n-models --local-dir ~/models/2D-to-3D-yolov11n-models
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of 2D-to-3D-yolov11n-models (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

linhjr87/2D-to-3D-yolov11n-models
Author
linhjr87
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1130

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments