Introduction:

In this project, we collect a large scale benchmark named WildFish for fish recognition in the wild. To our best knowledge, it is the largest fish dataset compared with existing fish datasets. It consists of 1,000 fish categories with 54,459 unconstraint images, enabling the training of high-capacity models for automatic fish classification. In addition, we also propose some novel open-set classification exploration practices and leverage pairwise textual descriptions to distinguish highly-confused species based on some realistic scenarios. More details can be found in our paper.

News:

• Please reach me at zpq0316@163.com, if you have difficulties when using BaiduYun to download the dataset.
• Nov. 12, 2018 We officially release WildFish_version1, which was used in our ACM-MM paper.

Tasks:

1. Open-Set Fish Classification:

2. Fine-grained Fish Recognition with Pairwise Text Descriptions

Download:

WildFish dataset can be downloaded from [BaiduYun]. All the settings in our paper can be found in the zip file.

Citing:

Please kindly cite the following paper, if you use WildFish in your work.

@inproceedings{zhuang2018wildfish,
  title={WildFish: A Large Benchmark for Fish Recognition in the Wild},
  author={Zhuang, Peiqin and Wang, Yali and Qiao, Yu},
  booktitle={2018 ACM Multimedia Conference on Multimedia Conference},
  pages={1301--1309},
  year={2018},
  organization={ACM}
}

Contact:

Please feel free to contact zpq0316@163.com, if you have any questions about WildFish.