HumanNeRF-SE: A Simple Yet Effective Approach to Animate HumanNeRF with Diverse Poses

We present HumanNeRF-SE, a simple yet effective method that synthesizesdiverse novel pose images with simple input. Previous HumanNeRF works require alarge number of optimizable parameters to fit the human images. Instead, wereload these approaches by combining explicit and implicit humanrepresentations to design both generalized rigid deformation and specificnon-rigid deformation. Our key insight is that explicit shape can reduce thesampling points used to fit implicit representation, and frozen blendingweights from SMPL constructing a generalized rigid deformation can effectivelyavoid overfitting and improve pose generalization performance. Our architectureinvolving both explicit and implicit representation is simple yet effective.Experiments demonstrate our model can synthesize images under arbitrary poseswith few-shot input and increase the speed of synthesizing images by 15 timesthrough a reduction in computational complexity without using any existingacceleration modules. Compared to the state-of-the-art HumanNeRF studies,HumanNeRF-SE achieves better performance with fewer learnable parameters andless training time.