GRF: Learning a General Radiance Field for 3D Representation and Rendering

Neural 3D Video Synthesis from Multi-view Video

“…There are several extensions of NeRF ] that are currently not peer reviewed. We recognize their efforts in improving neural radiance fields for in-the-wild scenes , generalization across scenes [Schwarz et al 2020;Trevithick and Yang 2020;, for non-rigid reconstruction [Du et al DyNeRF (r, g, b) < l a t e x i t s h a 1 _ b a s e 6 4 = " C k l 0 q s n e C / R y m 4…”

NeX: Real-time View Synthesis with Neural Basis Expansion

“…This method can handle view-dependent effects as the viewing angle is part of the 5D radiance function. Subsequent work improves upon NeRF by using explicit sparse voxel representation to improve fine detail (NSVF) [18], parameterizing the space to better support unbounded scenes (NeRF++) [50], incorporating learned 2D features that help enforce multiview consistency (GRF) [42], or extending NeRF to handle photometric variations and transient objects in internet photo collections (NeRF-W) [20]. Another related line of work involves implicitly modeling surface reflectance properties in addition to the scene geometry [1] or the light transport function [52].…”

“…This greatly limits the types of objects and scenes that MPI can capture. Recent research on implicit scene representation has made significant progress in the past months [21,32,18,50,42] and can be applied to view synthesis problem. Unfortunately, its expensive network inference still prohibits real-time rendering, and reproducing complex surface reflectance with high fidelity still remains a challenge.…”

Virtual Elastic Objects

“…Concurrently with steady progress towards realtime photo-realistic rendering of 3D environments in game engines, the last few decades have seen great strides towards photo-realistic 3D reconstruction. A recent achievement in this direction is the discovery of a fairly general formulation for representing radiance fields [5,27,31,32,36,45,46,49,53,61,63]. Neural radiance fields are remarkably versatile for reconstructing real-world objects with high-fidelity geometry and appearance.…”