Congrats @BenMildenhall, @_pratul_, and the entire NeRF team! It's difficult to highlight enough how much impact NeRF had on our research area.

However, I feel what's often overlooked how big of a jump this was in a single paper at the time: 1) differentiable rendering with a 3D representation that encodes density and appearance values 2) a neural field scene representation: combination of MLPs + positional encodings (they could've used a grid too but they didn't and it provided better local resolution). 3) view-dependent effects realized with view angle conditioning, regularized by feeding it into the network at a later stage (so less capacity for this part to mitigate overfitting).

The result was a new paradigm for joint 3D scene reconstruction and neural rendering with literally thousands of follow up papers - a once in a life-time work!