1 paper · 1 filter
Yiming Shu, Peng Sun, Lang Zhang +5
Feedforward 3D reconstruction has largely followed an {appearance-driven paradigm}: multi-view images are encoded into visual representations, cross-view correspondence is implicit…