Sam Hodge’s Post

View profile for Sam Hodge

Machine Learning Software Engineer and Computer Graphics Professional | BSc (Hons.), Dip Applied Design, Grad Dip (CompSci.) | Hollywood VFX Veteran

Visiting this dataset again from an lane in Prague, it reminds me how important bundle adjustment is during training of novel view synthesis models. Based on the slight lens and motion blur artifacts from a video shoot in 6 minutes of footage. The off the shelf products for camera intrinsic (lens) and extrinsics (pose) gave poor results. But if you optimise these camera parameters while you solve the novel view field things get better. Unfortunately getting these back out to real time models needs more work. But it is getting close to broadcast quality, from a still photograph session of about a 1-2 hour shoot with proper exposure control I think we can get there. #fisheye #NeRF #novelviewsynthesis #vfx #environments #VR #virtualproduction

To view or add a comment, sign in

Explore topics