A-Scan2BIM

Official implementation of the paper A-Scan2BIM: Assistive Scan to Building Information Modeling (BMVC 2023, oral)

Please also visit our project website for a video demo of our assistive system.

Updates

[06/14/2024] Updated our assistive demo so it accepts user-provided point cloud. Please visit here for instructions.

[12/08/2023] We are unable to release data for two floors, as the point clouds are unfortunately not available for download.

[11/30/2023] Due to a bug in our heuristic baseline method, we have updated our order evaluations. Please see the updated arXiv paper for the latest results.

Table of contents (also TODOs)

[x] Prerequisites
[x] Quickstart
[x] Training and evaluation
[ ] Collecting your own data
[ ] Building and developing the plugin
[x] Contact
[x] Bibtex
[x] Acknowledgment

Prerequisites

For more flexibility, our system is designed as a client-server model: a Revit plugin written in C# serves as the front end, and a python server which performs all the neural network computation.

To run our system, there are two options:

Run both the plugin and the server on the same machine
Run the server on a separate machine, and port-forward to client via local network or ssh

The advantage of option 2 over 1 is that the server can run on either Linux or Window machines, and can potentially serve multiple clients at the same time. Of course, the client machine needs to be Windows as Revit is Windows-only.

The code has been tested with Revit 2022. Other versions will likely work but are not verified. If you are a student or an educator, you can get it for free here.

Quickstart

To try out our assistive system, please visit here for instructions.

Training and evaluation

At a high-level, our system consists of two components: candidate wall enumeration network, and next wall prediction network. They are trained stand-alone, which we will describe the process step-by-step.

Data/environment preparation

First, fill out the data request form here. We will be in contact with you to provide the data download links. Once you have the data, unzip it in the root directory of the repository.

Since we do not own the point clouds, please download them from the workshop website. You would need the data from the 3D challenge, both the train and test data. Rename and move all the LAZ files to the data folder like below:

data/
├── history/
├── transforms/
├── all_floors.txt
└── laz/
    ├── 05_MedOffice_01_F2.laz
    ├── 06_MedOffice_02_F1.laz
    ├── 07_MedOffice_03_F3.laz
    ├── 08_ShortOffice_01_F2.laz
    ├── 11_MedOffice_05_F2.laz
    ├── 11_MedOffice_05_F4.laz
    ├── 19_MedOffice_07_F4.laz
    ├── 25_Parking_01_F1.laz
    ├── 32_ShortOffice_05_F1.laz
    ├── 32_ShortOffice_05_F2.laz
    ├── 32_ShortOffice_05_F3.laz
    ├── 33_SmallBuilding_03_F1.laz
    ├── 35_Lab_02_F1.laz
    └── 35_Lab_02_F2.laz

To install all python dependencies, run the provided script: sh ./setup_env.sh.

Then run the following command to preprocess the data:

cd code/preprocess
python data_gen.py

Also please download pretrained models from here and extract to the root directory of the repository. Some weights are necessary for network initialization (pretrained/) or evaluation (ae/), but others you may delete if training from scratch.

Training candidate wall enumerator

We borrow the HEAT architecture, which consists of two components: corner detector, and edge classifier. However in our case, we do not train end-to-end due to the large input size.

cd code/learn

# Step 1: train corner detector
for i in {0..15}
do
  python train_corner.py --test_idx i
done

# Step 2: cache detected corners to disk
python backend.py export_corners

# Step 3: train edge classifier
for i in {0..15}
do
  python train_edge.py --test_idx i
done

# Step 4: cache detected edges to disk
python backend.py save_edge_preds

Note that we do leave-one-out (per-floor) cross-validation, hence the for loops for step 1 and 3.

Training next wall predictor

Our next wall predictor along with the classifier baseline method are both trained on GT walls and order.

Training is initialized from pretrained weights of the previous candidate wall enumeration task (variant where only one reference point is used).

We have provided the pretrained weights for your convenience (see above for link to pretrained checkpoints), but they may not be necessary if training for your own task.

cd code/learn

# Training our method (metric-learning based)
for i in {0..15}
do
  python train_order_metric.py --test_idx i
done

# Training classifier baseline
for i in {0..15}
do
  python train_order_class.py --test_idx i
done

Evaluation

To evaluate reconstruction metrics:

python backend.py compute-metrics

To evaluate order metrics:

python eval_order.py eval-all-floors
python eval_order.py plot-seq-FID

To evaluate entropy and accuracy of next wall prediction:

python eval_order.py eval-entropy
python eval_order.py eval-all-acc-wrt-history

Collecting your own data

Coming soon...

Building and developing the plugin

Coming soon...

Contact

Weilian Song, weilians@sfu.ca

Bibtex

@article{weilian2023ascan2bim,
  author    = {Song, Weilian and Luo, Jieliang and Zhao, Dale and Fu, Yan and Cheng, Chin-Yi and Furukawa, Yasutaka},
  title     = {A-Scan2BIM: Assistive Scan to Building Information Modeling},
  journal   = {British Machine Vision Conference (BMVC)},
  year      = {2023},
}

Acknowledgment

This research is partially supported by NSERC Discovery Grants with Accelerator Supplements and the DND/NSERC Discovery Grant Supplement, NSERC Alliance Grants, and the John R. Evans Leaders Fund (JELF).

We are also grateful to the CV4AEC CVPR workshop for providing the point clouds.

And finally, much of the plugin code was borrowed from Jeremy Tammik's sample add-in WinTooltip. Hats off :tophat: to Jeremy for all the wonderful tutorials he has written over the years.

weiliansong / A-Scan2BIM

readme