Home
Catching up with newer research in image based rendering: A TLDR on how traditional computer graphics fits with neural rendering, computer vision, machine learning and capture hardware.
Why does it matter? This is likely the biggest change to graphics since the mid 80ies: We can now render lightfields, the holy grail of graphics. We know about light fields since 1936 but until 2.5 years ago they were just too expensive and impractical to create and render. You needed the camera on the picture below at least, now you render one from 30 frames of compressed video from youtube and a gaming machine or the cloud... In the space of 2 years NeRFs have gone from 12 hours to train a single scene and 30 minutes to render novel views, to training in 5 seconds and real-time rendering.
