3D human model creation on a serverless environment

Abstract

The creation of realistic 3D human model is traditionally timeconsuming and cumbersome, and is typically done by professionals. In recent years computer vision technologies can assist in generating human models from controlled environments, we demonstrate a different but easy capturing scenario with less constraints on the subject or the environmental setup. The reconstruction process for 3D human model consists of various intermediate process such as semantic human segmentation, human skeletal keypoint detection, and texture generation. In order to achieve easy, scalable, and flexible deployment to different cloud environments, we have chosen the serverless architecture to offload some common service functionalities to the cloud infrastructure but focused on the core task,which is the reconstruction itself. The event-driven serverless architecture eases the building of such multimedia web services with minimal coding efforts, but simply defines the APIs and declares the APIs with correspondent lambda functions. The proposed approach in this paper allow anyone with a mobile phone to generate 3D models easily and quickly in the scale of few 2-3 minutes, rather than hours.