DT-LSD: Deformable Transformer-based Line Segment Detection

This repository contains the official implementation for DT-LSD (Deformable Transformer-based Line Segment Detector)

TODO

• Training
• Evaluation
• Upload source code
• Upload weight
• Inference
• Upload arxiv paper

Results

Dataset	sAP10	sAP15	APH	FH	FPS
Wireframe	71.7	73.9	89.1	85.8	8.9
YorkUrban	33.2	35.1	65.9	68.0	8.9

Installation

1. Clone this repository.

   git clone https://github.com/SebastianJanampa/DTLSD.git
   cd DTLSD

2. Install Pytorch and torchvision

   Follow the instructions on https://pytorch.org/get-started/locally/.

   # an example:
   conda install -c pytorch pytorch torchvision

3. Install other needed packages

   pip install -r requirements.txt

4. Compiling CUDA operators

   cd models/dtlsd/ops
   python setup.py build install
   # unit test (should see all checking is True)
   python test.py
   cd ../../..

Dataset

To reproduce our results, you need to process two datasets, ShanghaiTech and YorkUrban.

mkdir data
cd data
wget https://github.com/SebastianJanampa/storage/releases/download/v1.0.0/wireframe_processed.zip
wget https://github.com/SebastianJanampa/storage/releases/download/v1.0.0/york_processed.zip

unzip wireframe_processed.zip
unzip york_processed.zip

rm *zip
cd ..

Pretraining weights

Download the weights from DINO_SWIN_4scales_36_epochs from the DINO repo, and place it in the pretrain folder.

Run

1. Training

bash scripts/train/DTLSD_SWIN_4_scales_24_epochs.sh 

2. Testing

bash scripts/train/DTLSD_SWIN_4_scales_24_epochs.sh

Demo

Download the DTLSD weights

wget https://github.com/SebastianJanampa/storage/releases/download/v1.0.0/DTLSD_checkpoint0035.pth

If wget doesn't work, download it from this link, and place it in the main folder. Then run

python demo.py

Citation

@article{janampa2024dt,
  title={DT-LSD: Deformable Transformer-based Line Segment Detection},
  author={Janampa, Sebastian and Pattichis, Marios},
  journal={arXiv preprint arXiv:2411.13005},
  year={2024}
}