Structured Outdoor Architecture Reconstruction by Exploration and Classification

Fuyang Zhang, Xiang Xu, Nelson Nauata, Yasutaka Furukawa.

[arXiv] [Project Page] [Bibtex]

In ICCV 2021

Prerequisites

• Linux
• NVIDIA GPU, CUDA 11+
• Python 3.7+, PyTorch 1.7+

Dependencies

Install additional dependencies:

$ pip install -r requirements.txt

Data

Download the processed data from this link. This includes the original cities dataset from "Vectorizing World Buildings: Planar Graph Reconstruction by Primitive Detection and Relationship Classification" and predictions from Conv-MPN, IP and Per-Edge models.

Download the pretrained heatmap weights from this link.

Both data are required for training and evaluation, unzip and move them to the data folder.

Running the Code

Training

python train_evaluators.py

This will start both the train and search threads.

You can change settings like beam search depth or number of training epochs in the config.py.

Evaluation

First, perform beam search over all the test data:

python search_result.py

Then, evaluate the scores for all searched results:

python metric_for_result.py

Pretrained models

Download individual pretrained model and its beam search results.

Training Dataset	Model Weights	Beam Search Results
Conv-MPN	convmpn_weights.zip	convmpn_beamsearch.zip
IP	ip_weights.zip	ip_beamsearch.zip
Per-Edge	peredge_weights.zip	peredge_beamsearch.zip

Citation

If you find this code helpful, please consider citing:

@InProceedings{zhang2021structured,
      title={Structured Outdoor Architecture Reconstruction by Exploration and Classification}, 
      author={Fuyang Zhang and Xiang Xu and Nelson Nauata and Yasutaka Furukawa},
      year={2021},
      eprint={2108.07990},
      archivePrefix={International Conference on Computer Vision (ICCV)},
      primaryClass={cs.CV}
}

Contact

If you have any questions, please contact [email protected] or [email protected]

Acknowledgement

This research is partially supported by NSERC Discovery Grants with Accelerator Supplements and DND/NSERC Discovery Grant Supplement.