The human head was based on a scanbased commercial avatar with blendshape animation, exploiting a multichannel texture simulation software.Mesh deformations compatible with the human face muscle structure were computed from motion capture data in the same way as for the monkey face.The used dynamic head models achieve stateoftheart degree of realism for the human head, and to our knowledge, we present the only highly realistic monkey avatar that is animated with motion capture data from real animals used in physiology so far.It has been demonstrated by a recent study of our lab that our dynamic monkey avatar induces behavioral reactions of macaque monkeys that are very similar to ones elicited by real movies, reaching the good side of the uncanny valley, contrasting with previous studies using avatars in experiments with monkeys. A related result has been obtained recently for static pictures of monkeys, demonstrating comparable looking times for the avatar and real pictures of monkey expressions, but without expressive motion of the face. For motion capture, we used a VICON motion capture system with a marker set of markers that were placed on the face of a monkey and a human participant.Facial expressions were elicited by instructions, or by having the animal interact with an experimenter, respectively.For this study, we exploited multiple repetitions of two human and two monkey expressions, and additional trials with neutral expressions.In order to control the information content and the expressivity of the dynamic face stimuli, we created motion morphs between these <a href="https://www.targetmol.com/compound/Pivanex">buyPivanex</a> prototypical expressions.This algorithm allows to create linear combinations in spacetime between these prototypical motions, controlling smoothly the expressiveness and the style of the created facial motion.The different versions of the model were compared exploiting their prediction accuracy and the BIC.We discarded the models if, after addition of a new predictor, either their accuracy was decreasing or the BIC showed a decrease.The tuning function to monkey expressions as a function of the species parameterswas defined as DM s R larly, the tuning function for human expressions was given by DH s R de.For this purpose, we first defined different measures for the lowlevel information content and balanced the stimuli by equilibrating these measures.Tested measures included, the maximum amount of deformation, and the. To control the information content of the stimuli, we generated morphs between the original motion and the trajectories of a neutral expression using our motion morphing technique.In these morphs, the original expression was weighted with the morph levelland the neutral expression with the weight l. The parameterlwas chosen to equate the lowlevel measures of all four prototypical stimuli, <a href="https://www.ncbi.nlm.nih.gov/pubmed/17267032">searchPivanex</a> separately for the two avatar models. The MF measure resulted in the least variability of the perceived expressiveness of the equilibrated stimuli, and thus was used to equilibrate the stimuli for all experimental conditions.Models of different complexity were compared based on the prediction accuracy and by exploiting the BIC.Two statistical measures were applied in order to compare the similarity of the categorization responses for the two avatar types.As second statistical analysis, we compared the multinomially distributed fourclass classification responses across the participants for the individual points in morphing space using a contingency table analysis that tested for the independence of the class probabilities from the avatar types and the two view conditions.
