Learning Attentive Pairwise Interaction for Fine-Grained Classification (API-Net)

Peiqin Zhuang, Yali Wang, Yu Qiao

Introduction:

In order to effectively identify contrastive clues among highly-confused categories, we propose a simple but effective Attentive Pairwise Interaction Network (API-Net), which can progressively recognize a pair of fine-grained images by interaction. We aim at learning a mutual vector first to capture semantic differences in the input pair, and then comparing this mutual vector with individual vectors to highlight their semantic differences respectively. Besides, we also introduce a score-ranking regularization to promote the priorities of these features. For more details, please refer to our paper.

Framework:

Dependencies:

Python 2.7
Pytorch 0.4.1
torchvision 0.2.0

How to use:

# python train.py

Citing:

Please kindly cite the following paper, if you find this code helpful in your work.

@inproceedings{zhuang2020learning,
  title={Learning Attentive Pairwise Interaction for Fine-Grained Classification.},
  author={Zhuang, Peiqin and Wang, Yali and Qiao, Yu},
  booktitle={AAAI},
  pages={13130--13137},
  year={2020}
}

Contact:

Please feel free to contact zpq0316@163.com or {yl.wang, yu.qiao}@siat.ac.cn, if you have any questions.

Acknowledgement:

Some of the codes are borrowed from siamese-triplet and triplet-reid-pytorch. Many thanks to them.

inSight-mk1/API-Net