Scaling images causes aliasing

[RuudBurger]

What would be the best option to scale images? I understand when I scale up it would impact quality, but I want the image at least be aliased better.

Original:

Scaled:

The antiAlias prop doesn't seem to do anything for this.

I'm basically doing this.

<Group transform={[{ scale: 0.6 }]}>
   <Image fit="contain" height={500} image={image} width={500} x={0} y={0} />
</Group>

[wcandillon]

@RuudBurger Thank you for getting in touch again, I will try the tvOS PR this week. You are bringing an interesting and timely point. I believe that this issue is due to the default sampling option. This relates to #2509.
Now I had left this issue aside because there some architectural improvements I wanted to do first (especially #2865 without which such a change would be unnecessarily tedious). Now it's time to fix this 100%

As a workaround, using the imperative API, you have more (full? I will need to check) control over the sampling options:

[RuudBurger]

Thanks again for the quick answer.
I just patched our current build with the SkSamplingOptions({0, 0}) change from the linked PR. For our current use-case I think it's fine to always use the cubic sampling.

[wcandillon]

I have made a PR for it: #2880 . Could you send me the image and some scaling I should do it? I could add it to the test suite.

[RuudBurger]

This has the one I showed you earlier. It works better then the Moana image above

but any image with straight diagonal lines would do.

Then scale it up by 25% in skia so it's 1600px on screen.

[wcandillon]

thks. I did a bit more testing, while cubic sampling will yield the best result, the issue is that we were defaulting to the wrong sampling (nearest instead of linear). Linear will yield a descent result as well.

[RuudBurger]

Any big performance differences?

[wcandillon]

on paper cubic is slower but that would be only relevant if you would sample tons of images for a single frame. In your case, I don't think you could feel the difference.

…

On Sat, Jan 11, 2025 at 10:29 AM Ruud Burger ***@***.***> wrote: Any big performance differences? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>