Resize node creates artefacts in ACES CG!

Hi all.

Working on a project in ACES.
Elements come in ACES 2065, we convert them either in ACES CG or ACES CCT when needed.

We have an unexpected thing with Resize node.
When used on ACES CG images, it sometimes creates ugly artefacts in the white parts of the images. Like clamped portions.
If we convert to ACES CCT, do the resize and come back to CG, it looks perfect.
Is that a bug of the resize node or is there a reason why it does such things (that we might not notice)?

Tes, have a good flame day

It`s an expected behavior. Filtering can create negative results in Linear. Resize in log space (ACEScct is log) is a way to fix it.

7 Likes

This. :arrow_up:

1 Like

Certain resize filtering types, especially Lanczos, are peculiar in that they can produce negative numbers with non-negative data sets.

Shameless self promotion for the ACES class on Logik Academy Pro that goes into detail of why this happens and how to avoid it.

3 Likes

Shame is on me for being so ignorant about all that scary color management traps…
Thx

2 Likes

Please don’t shame yourself. I’d reckon less than 5% of the members of our community are aware of this. Plus, some of us just aren’t right in the head. I mean, I read Wikipedia articles on Lanczos.

Oh My God Omg GIF by This Is Us

2 Likes

“Since the kernel assumes negative values for a > 1, the interpolated signal can be negative even if all samples are positive.”

2 Likes

How does one correctly pronounce Lanczos?

2 Likes

With a French accent… and a reverence.

5 Likes

I’ve gone down this rabbit hole before! He was Hungarian, Cornelius Lanczos, and based on the hungarian phonetic alphabet spelling on his wikipedia page, it appears to me that lahn-tzōsh might be a close approximation.

2 Likes

Haha … just came in here to say Lanczos was Hungarian

3 Likes

Edit: ignore me.

3 Likes

Pronunciation would be like: Laaanchosh (with a long “a”)

4 Likes

Can’t

2 Likes

Can

1 Like

Resize on the timeline with mitchell filtering gets rid of these artifacts most of the time. It’s marginally softer yes and the cct way is of course the correct way. But mitchell will do it in a pinch