Appearance-based Gaze Estimation with Deep Learning: A Review and Benchmark.

Human gaze provides valuable information on human focus and intentions, making it a crucial area of research. Recently, deep learning has revolutionized appearance-based gaze estimation. However, due to the unique features of gaze estimation research, such as the unfair comparison between 2D gaze positions and 3D gaze vectors and the different pre-processing and post-processing methods, there is a lack of a definitive guideline for developing deep learning-based gaze estimation algorithms. In this paper, we present a systematic review of the appearance-based gaze estimation methods using deep learning. Firstly, we survey the existing gaze estimation algorithms along the typical gaze estimation pipeline: deep feature extraction, deep learning model design, personal calibration and platforms. Secondly, to fairly compare the performance of different approaches, we summarize the data pre-processing and post-processing methods, including face/eye detection, data rectification, 2D/3D gaze conversion and gaze origin conversion. Finally, we set up a comprehensive benchmark for deep learning-based gaze estimation. We characterize all the public datasets and provide the source code of typical gaze estimation algorithms. This paper serves not only as a reference to develop deep learning-based gaze estimation methods, but also a guideline for future gaze estimation research. The project web page can be found at https://phi-ai.buaa.edu.cn/Gazehub/.

Morphological characteristics influence the spatial mixing patterns of shorebirds at Shengjin Lake.

The coexistence of species with similar ecological niches is one of the core interests of community ecology research. However, how functional feeding traits, including bill size and leg length, determine the niche of mixed flocks of shorebird species has seldomly been studied, as well as, microhabitat variables affect the spatial patterns of availability and the quality of patches for wintering. From October 2016 to March 2017 at Shengjin Lake, Anhui Province, China, we recorded 226 scan samples from the different microhabitats and 93 focal animal videos of four common shorebird species: common greenshank, spotted redshank, Kentish plover, and little ringed plover. We found that the species participating in the mixed groups were different in each microhabitat. The results of the overlap index for microhabitats and foraging techniques between the species were consistent with the morphological characteristics of these species. Kentish and little ringed plovers had the highest Pianka's niche overlap index values of 0.95 and 0.98 for microhabitats and foraging techniques, respectively, whereas common greenshank and spotted redshank had values of 0.78 and 0.89, respectively. Common greenshank and spotted redshank used four foraging techniques: a single probe (PR), multiple probes (MPR), a single peck (PE), and multiple pecks (MPE). Kentish and little ringed plovers only used PE and MPE. The mean bill size, mean leg length, and mean foraging frequency were significantly associated with water depth. The mean bill size and mean leg length were both significantly correlated with the mean foraging frequency of shorebirds. The vegetated area was the most important variable for grouping among shorebirds. We concluded that the four species showed differences in their preferred microhabitats and foraging patterns. Interspecific morphological differences, including bill and leg lengths, resulted in niche differentiation. Thus, effective resource allocation by regional species was realized, and a dynamic balance was achieved by the mixed foraging species. The information on foraging behavior and habitat requirements could be useful in the management of water levels in natural areas and conservation of a diversity of wintering shorebirds.

Effect of conspecific neighbors on the foraging activity levels of the wintering Oriental Storks (Ciconia boyciana): Benefits of social information.

Animals prefer to aggregate in patches with high abundance and availability of food resources. Group foragers typically receive information about food resources by monitoring external events and the behavior of neighbors. The Information Centre Hypothesis proposes that aggregations increase foraging activity levels as a result of social information provided by conspecifics. Increasing the foraging rate has as a result decreasing time devoted to anti-predator vigilance and may intensify competition among group members. Studies have shown that foraging activities are influenced by factors other than flock size, such as the number and foraging intensity of neighbors. To test these hypotheses, we examined the effect of number and foraging intensity of neighbors on the foraging activity levels (foraging rate, foraging effort, and foraging success rate) of the wintering Oriental Storks (Ciconia boyciana). In this study, we collected focal sampling data on the foraging behavior of storks at Shengjin Lake during winter from 2017 to 2019, controlling the effects of other variables (group identity, wintering years, and wintering periods). We found that foraging activity levels were higher in the presence of foraging neighbors than in their absence. Moreover, individuals adjusted their foraging activity levels according to social information gathered from the behavior of neighboring conspecifics. Focal individuals' foraging rate and foraging effort were positively correlated with the average foraging rate of neighbors. Their foraging success rate was not influenced by the average foraging rate and foraging success rate of neighbors; however, it was positively correlated with the average foraging effort of neighbors. In conclusion, foraging activity levels of individuals are primarily driven by the intensity of the foraging activity of neighbors. This result differs from the results of previous studies that suggested that flock size was the most important factor determining individual foraging activity levels.