Canonical Fields: Self-Supervised Learning of Pose-Canonicalized Neural Fields

cs.brown.edu/projects/canonicalfields Figure 1 . We present Canonical Field Network (CaFi-Net), a self-supervised method for 3D position and orientation (pose) canonicalization of objects represented as neural fields. We specifically focus on neural radiance fields (NeRFs) fitted to raw RGB images of arbitrarily posed objects (left), and obtain a canonical field (fixed novel view shown on right) where all objects in a category are consistently positioned and oriented.