MoCo‐Flow: Neural Motion Consensus Flow for Dynamic Humans in Stationary Monocular Cameras

Synthesizing novel views of dynamic humans from stationary monocular cameras is a specialized but desirable setup. This particularly attractive as it does not require static scenes, controlled environments, or capture hardware. In contrast to techniques that exploit multi-view observations, the problem modeling scene single view significantly more under-constrained and ill-posed. this paper, we introduce Neural Motion Consensus Flow (MoCo-Flow), representation models in using 4D continuous time-variant function. We learn proposed by optimizing for minimizes total rendering error, over all observed images. At heart our work lies carefully designed optimization scheme, which includes dedicated initialization step constrained motion consensus regularization on estimated flow. extensively evaluate MoCo-Flow several datasets contain human motions varying complexity, compare, both qualitatively quantitatively, baselines ablated variations methods, showing efficacy merits approach. Pretrained model, code, data will be released research purposes upon paper acceptance.