[ECCV 2020] Semantic Line Detection Using Mirror Attention and Comparative Ranking and Matching

Dongkwon Jin, Jun-Tae Lee, and Chang-Su Kim

Official implementation for "Semantic Line Detection Using Mirror Attention and Comparative Ranking and Matching" [paper] [supp] [video] [arxiv].

Source code for baseline method (SLNet) is available in here.

Also, more recent work can be found in here.

Video

Requirements

• PyTorch 1.3.1
• CUDA 10.0
• CuDNN 7.6.5
• python 3.6

Installation

Create conda environment:

    $ conda create -n DRM python=3.6 anaconda
    $ conda activate DRM
    $ pip install opencv-python==3.4.2.16
    $ conda install pytorch==1.3.1 torchvision cudatoolkit=10.0 -c pytorch

Download repository:

    $ git clone https://github.com/dongkwonjin/Semantic-Line-DRM.git

Instruction

1. Download the following datasets to root/. SEL and SEL_Hard are datasets for semantic line detection. Others are datasets for applications. We obtain the edge detection results in edge folder, by employing HED algorithm.

Dataset	Custom	Original path
SEL	Download	here
SEL_Hard	Download
AVA landscape	Download	here
ICCV	Download	here
NYU	Download	here
SYM_Hard	Download

2. Download our model parameters to root/(task_folder_name)/ if you want to get the performance of the paper.

Task	Model parameters
Semantic line detection	Download
Dominant parallel line detection	Download
Reflection symmetry axis detection	Download

3. Edit config.py. Please modify dataset_dir and paper_weight_dir. If you want to get the performance of the paper, please input run_mode to 'test_paper'.

4. Run with

cd Semantic-Line-DRM-master/(task_folder_name)/(model_folder_name)/code/
python main.py

Reference

@Inproceedings{
    Jin2020DRM,
    title={Semantic Line Detection Using Mirror Attention and Comparative Ranking and Matching},
    author={Dongkwon Jin, Jun-Tae Lee, and Chang-Su Kim},
    booktitle={ECCV},
    year={2020}
}