Render filters

Started by sboerner, January 26, 2021, 10:46:28 AM

Previous topic - Next topic

sboerner

As a rule I use the Mitchell-Netravali filter and have been pleased with the results, but lately I've run into a scene with a lot of distant, detailed foliage and am experimenting to see if one of the others might work better.

Just curious what everyone's thoughts might be. Do you always use the same filter, or use different filters for different types of scenes?

Very few threads on this topic, it seems. One that looked promising refers to a broken link.

RichTwo

I Always without thinking use the default Catmull-Rom, and never knew what effect any other filter would have.  It'd be great if it'd cut back render times, say for doing path tracing.  I'm on a project with PT rendering and it's taking over three hours... ugh!  :P  

I do see there's "Box" filter "Not recommended"?  Now why would such a thing be an option?  That's like keeping something poisonous in the food pantry...
They're all wasted!

WAS

I use soft, cause edges can be too hard and have outlines sometimes. Cause then I can just sharpen later. And because it had no edge correction I can stabilize small particles in motion like my star shader.

Tangled-Universe

My preference is MN, but lately I have switched most often to CR. It's perhaps a tad bit too sharp, but with AA8 it looks fine and because I always use deferred shading I seldomly render lower than AA8.

sboerner

Thanks for all the responses.

Soft makes sense for animation. I'll be trying Catmull-Rom; I don't think I've ever used that one. The scene I'm working on has a large field of grass, and the camera angle is high and distant. I'd like to keep as many edges as I can.

From what I recall about filter patterns (which isn't much :) ), the box filter is about as rudimentary as it gets. Probably good advice to steer clear.

Dune

Actually I often forget to change the filter, and use default, for experiments and 'fun' renders at least. For commissioned work and final renders I do use MN a lot.

bobbystahr

Quote from: RichTwo on January 26, 2021, 11:22:36 AMI do see there's "Box" filter "Not recommended"?  Now why would such a thing be an option?  That's like keeping something poisonous in the food pantry...
agree, most curious....
something borrowed,
something Blue.
Ring out the Old.
Bring in the New
Bobby Stahr, Paracosmologist

Kadri

It might be different now, i don't know. But when we tested some noise filters ( i think it was a nvidia denoiser) last year
some results were better with the Terragen old box filter.
It is old and not the best but who knows when something like that happens. It was interesting to see those results.

zaxxon

I used to use the MN filter almost exclusively, I recall Matt saying something to the effect that it "handled pixels really well". I didn't really understand what he meant (still don't), but I just defaulted to it. However as I added more dense and detailed populations I really found that I preferred the way the detail looked with the CR filter, and when necessary used a bit of blur in post. The 'softer' filters never gave me the look I wanted and sharpening in post was often problematic. It may be subjective but the CR filter with the Path Tracer seems to give even better results.

WAS

Quote from: zaxxon on January 28, 2021, 06:36:27 PMThe 'softer' filters never gave me the look I wanted and sharpening in post was often problematic
You need raw/32bit for good pixel reconstruction in post similar to normal photography (and film), and sharpening you'd need to use algorithm based, like smart sharpen and dial in Gaussian settings that match the softness spread of edges in the image. It is the best method allowing the the most flexible post in film and such for CGI, so you're not trying to match sharpness, and can just do it all in post and have a uniform look.

It's also painfully obvious in older films mixing traditional film and CGI, where film grain and shutter lighting was completely missing, let alone the CGI itself over-sharpened by contrast to the film.