A new dataset of dog breed images and a benchmark for finegrained classification

  • PDF / 5,170,642 Bytes
  • 11 Pages / 612 x 808 pts Page_size
  • 35 Downloads / 299 Views

DOWNLOAD

REPORT


A new dataset of dog breed images and a benchmark for finegrained classification Ding-Nan Zou1,2 , Song-Hai Zhang1 (

), Tai-Jiang Mu1 , and Min Zhang3

c The Author(s) 2020. 

to a need for dog identification using modern visual technology, both for dog recognition and finer-grained classification to breed. Fine-grained classification is a non-trivial problem, requiring to distinguish different subclasses from subtle inter-class differences. As for other visual tasks, the performance of fine-grained classification has been greatly boosted by the use of deep neural networks [1–4]. However, there are relatively small differences between dogs of different breeds while there can be relatively large differences between those within a breed due to geographic isolation or hybridization. See, for example, Fig. 1: great Dane dogs have multiple colors, while dogs of different breeds, such as Norwich terriers and Australian terriers, may have similar colors. Existing datasets, such as the widely used Stanford Dogs Dataset [5], are not diverse enough to cover such variations, limiting their use for training and testing algorithms. Keywords fine-grained classification; dog; dataset; This paper contributes a new dataset, Tsinghua benchmark Dogs, with an emphasis on fine-grained dog classification. It contains 130 breeds of dogs in 70,428 images, with one dog per image, over 65% of which 1 Introduction were collected from everyday life. It covers nearly Dogs are closely involved in human lives as family all dog breeds currently found in China. Each breed members, and are very common as pets. On the other in our dataset contains at least 200 images, up to a hand, the number of dog-related incidents of injury maximum of 7449 images, basically in proportion and uncivilized behavior is increasing. This leads to their frequency of occurrence in China, so it significantly increases the diversity for each breed over 1 Department of Computer Science and Technology, existing datasets. Furthermore, we have annotated BNRist, Tsinghua University, Beijing 100084, China. bounding boxes of the dog’s whole body and head E-mail: D.-N. Zou, [email protected]; in each image, which can be used for supervising S.-H. Zhang, [email protected] ( ); T.-J. Mu, the training of learning algorithms as well as testing [email protected]. them. 2 NaJiu Company, Hunan 410022, China. We have also benchmarked several classification 3 Harvard Medical School, Brigham and Women’s Hospital, methods on our dataset, including both general neural Boston, MA 02115, USA. E-mail: [email protected]. networks and fine-grained models which exhibit Manuscript received: 2020-05-18; accepted: 2020-06-14

Abstract In this paper, we introduce an image dataset for fine-grained classification of dog breeds: the Tsinghua Dogs Dataset. It is currently the largest dataset for fine-grained classification of dogs, including 130 dog breeds and 70,428 real-world images. It has only one dog in each image and provides annotated bounding boxes for the whole body and head. In comparison to p