Martin A. Giese¹, Lucas Martini¹, Alexander Lappe¹; ¹Hertie Institute / CIN, University Clinic Tuebingen, Germany

The neural and computational mechanisms of the visual encoding of body pose and motion remain poorly understood. One important obstacle in their investigation is the generation of highly controlled stimuli with exactly specified form and motion parameters. Avatars are ideal for this purpose, but for nonhuman species the generation of appropriate motion and shape data is extremely costly, where video-based methods often are not accurate enough to generate convincing 3D animations with highly specified parameters. METHODS: Based on a photorealistic 3D model for macaque monkeys, which we have developed recently, we propose a method that adjusts this model automatically to other nonhuman primate shapes, requiring only a small number of photographs and hand-labeled keypoints for that species. The resulting 3D model allows to generate highly realistic animations with different primate species, combining the same motion with different body shapes. Our method is based on an algorithm that deforms a polygon mesh of a macaque model with 10,632 vertices with an underlying rig of 115 joints automatically, matching the silhouettes of the animals and a small number of specified key points in the example pictures. Optimization is based on a composite error function that integrates terms for matching quality of the silhouettes, keypoints, and bone length, and for minimizing local surface deformation. RRSULTS: We demonstrate the efficiency of the method for several monkey and ape species. In addition, we are presently investigating in a psychophysical experiment how the body shape of different primate species interacts with the categorization of body movements of humans and non-human primates in human perception. CONCLUSION: Using modern computer graphics methods, highly realistic and well-controlled body motion stimuli can be generated from small numbers of photographs, allowing to study how species-specific motion and body shape interact in visual body motion perception.

Acknowledgements: ERC 2019-SyG-RELEVANCE-856495; SSTeP-KiZ BMG: ZMWI1-2520DAT700. Max Planck Research School for Intelligent Systems Tübingen.