setting a different halton sequence base set, 5 and 7, for the depth-of-field sampling doesnt seem to have too much of an effect, but it does change some of the artifacts

Tons of samples behind the focal plane

Much higher sample count, 32 to 256 adaptive with a more lenient error threshold. Much better results

Some hyperparam tuning to better resolve objects in the foreground. Still lots of noise in the heavily aliased sections...

First attempt without lecture, thin-lens approximation, dof as degrees for a deformation disk technique. Adaptive sampling with 16-64 halton sequence anti-aliasingLots of noise in the background and some blur in the foreground. Probably just needs some hyperparam tuning for that!