Downscaling debate

[radiant]

Hello guys,

Me and my "know it all" friend was having a debate about the point in rendering 4k images and downscaling them to 1k or a standered resolution.
The question is, is there any point or any difference shown at rendering 4k and downscaling to 1k then to just render a 1k image.

Please be detailed in your answers, he is the kind of guy that always thinks he is always right

Radiant

[steveps3]

Down scaling is a good way of reducing noise and increasing sharpness. Other than that, no real point.

[t_3]

Down scaling is a good way of reducing noise and increasing sharpness. Other than that, no real point.

... and reducing noise and increasing sharpness would (or could) mean, that there is more detail in the end than if it was rendered in 1k straight - at least if a scene holds enough details. for lq textures and low polys simple filtering would pretty much give the same effect.

[Lyubomir]

In Indigo RT, the supersampling factor does exactly that - it makes the renderer render at higher resolution and automatically downscales it for the display buffer and final saved image. By bumping this number to 10 I can completely get rid of any firefiles and the output becomes butter smooth without much effect on actual render speed. That would be very good to have in Octane.

[3form]

+ the SS is really good

[radiance]

Downscaling support in renderer is a dirty solution imo,
It basically relates to the engine not having a good reconstruction filter in the core when splatting samples onto a film buffer.

There are many different ways to tackle this problem, and many are only suitable for particular types of use.

Personally, i don't see supersampling support in octane to be a good option, the interactivity and GPU memory use of a 4000x4000 render buffer is FAR worse then at an original render resolution.

Then again, as most people always use some image processing app like photoshop or gimp to do final image adjustments, you can just render to 4000x4000 and then downscale in photoshop as you please.

Our priorities are trying to fix the issue propperly, by programming good reconstruction / filtering code.
We could definately add support for supersampling, it's a pretty basic feature to add, but again, if you read the above, you'll realise why it's not on the #1 priority spot right now...

Radiance