jingweiqu / eagm

[IJCAI'21] Adaptive Edge Attention for Graph Matching with Outliers

Adaptive Edge Attention for Graph Matching with Outliers

Aeroplane	Cow	Train

This repository is the implementation of the paper:

Jingwei Qu, Haibin Ling, Chenrui Zhang, Xiaoqing Lyu, and Zhi Tang. Adaptive Edge Attention for Graph Matching with Outliers. IJCAI, 2021.

It contains the training and evaluation procedures for the three benchmarks in the paper:

• Pascal VOC with Berkeley annotations
• Willow Object
• CMU House Sequence

Requirements

• Graph Nets (TensorFlow 1 and Sonnet 1)
• TensorFlow (>= 1.2.0)

Dataset

Download and prepare the datasets:

sh Data_Download.sh

Willow Object, CMU House Sequence, and Berkeley annotations of Pascal VOC are provided in the folder data.

Evaluation

Run evaluation using the trained models provided in the folder trained_models:

sh Test_dataset.sh

where dataset is the benchmark to be evaluated (PascalVOC, Willow, or CMUHouse).

Training

Run training:

sh Train_dataset.sh

Citation

@inproceedings{qu2021adaptive,
  title={Adaptive Edge Attention for Graph Matching with Outliers},
  author={Qu, Jingwei and Ling, Haibin and Zhang, Chenrui and Lyu, Xiaoqing and Tang, Zhi},
  booktitle={Proceedings of the International Joint Conference on Artificial Intelligence},
  pages={966--972},
  year={2021},
  publisher={IJCAI},
  doi={10.24963/ijcai.2021/134}
}