In this work, we demonstrated that receptive fields in the state-of-the-art 3D human pose estimation model by Facebook AI could be effectively specified using Optical Flow. We introduced adaptive receptive fields, a simple and effective method to aid receptive field selection in pose estimation models based on optical flow inference. We contrasted the performance of a state-of-the-art benchmark model running on fixed receptive fields with their adaptive field counterparts. When using a reduced receptive field, our model processed slow-motion sequences (10x longer) 23% faster than the benchmark model running at regular speed. The reduction in computational cost was achieved while producing a pose prediction accuracy within 0.36% of the benchmark model [arXiv].

In this project, we report on an experiment with The Walking Dead, a narrative-driven adventure game where players must survive in a post-apocalyptic world filled with zombies. We used OpenFace computer vision software to extract action unit (AU) intensities of facial expressions characteristic of decision-making processes taken during the game. We compared the AUs of two conditions: moral decisions and non-moral decisions, and demonstrated that the AU05 and AU45, which correspond to the eyelid raiser and the eye blink, respectively, could be involved during morally charged decision-making. Furthermore, we saw that pre-decision variations in the AU17 (chin raiser), AU23 (lip tightener), and AU25 (lips part) could be predictive of decision-making processes. This study was published at the Advances in Computer Games (ACG 2021) conference.