Learning Deep Features for Giant Panda Gender Classification using Face Images

Hongnian Wang, Han Su, Peng Chen, Rong Hou, Zhihe Zhang, Weiyi Xie; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

Giant panda (panda) has lived on earth for at least eight million years and is known as the living fossil. It is also a vulnerable species which requires urgent protection. It is essential to conduct population survey collecting information of their population, density, age structure, and gender ratio so as to design protection schemes and measure their effectiveness. However, it is challenging to accurately and timely obtain gender ratio of pandas because their pelage lacks distinguishable gender patterns and panda is sparsely distributed population in large habitats. All current approaches rely heavily on manual collection of samples in the wild, which are time consuming, costly, or even dangerous. With the widely deployed camera traps, if the gender of pandas can be determined from images, it is possible to monitor panda gender ratio in different regions in real-time. However, no such study was done. In this paper, a deep learning method is developed to study the distinctiveness of panda face for gender classification, in which the largest panda image dataset with 6,549 panda face images collected from 100 male and 121 female pandas is established. The experimental results show that panda faces contain some gender information, although they look very similar to human vision.