GHNeRF can simultaneously learn neural radiance fields and generalizable human features, such as 2D/3D joint locations and dense poses, from sparse 2D images.