Danbooru

JPEG artefacts in samples

Posted under General

Pretty sure he knows that from the "artefact...(mainly US: 'artifact')" in the article he posted.

Back to the topic at hand, I'm not sure if it's my eyes or my monitor, but you've always been more sensitive to JPEG artifacts than I've been. Do you notice this change only with samples?

If not, and if nothing was changed on Danbooru, I wonder if an external source (e.g. pixiv) might have tweaked the way they store or serve images (resampling to save on bandwidth). An image going through two levels of resampling is going to look much worse than one that only goes through one.

DOCoSPADEo said:
To be fair Haduki, "artefact" is the british variant of the word "artifact".

Why are you telling me this? I know that, that's why I use it. Besides, my name's Hazuki, and only Fencedude and LaC are allowed to spell it Haduki (due to them being widely known to be complete butts).

Shinjidude said:
Back to the topic at hand, I'm not sure if it's my eyes or my monitor, but you've always been more sensitive to JPEG artifacts than I've been. Do you notice this change only with samples?

If not, and if nothing was changed on Danbooru, I wonder if an external source (e.g. pixiv) might have tweaked the way they store or serve images (resampling to save on bandwidth). An image going through two levels of resampling is going to look much worse than one that only goes through one.

Good question. I've seen some pretty JPEG'd pics coming from pixiv, but then I've always had a very low opinion of the Japanese artists' ability not to ruin their own artwork. I can't honestly say if I see more JPEG in the originals or not, since I don't necessarily look at each picture in non-resampled size.

It'd be fairly interesting to run that double-compression test on recent uploads and compare it with older ones, but I don't think anyone's ever published working code to accompany the paper.

The samples are set to 90% quality, which I think is reasonable. Quality sticklers would probably use original images anyway.

That being said I'm not entirely happy with the resizer. It's definitely possible to write something more memory and processor efficient, but mucking around with C++ code isn't the easiest thing.

Right now the resizer is basically a wrapper around libgd. If you want to suggest a better library I'm open to suggestions.

I was more curious about the sudden drop in quality I've noticed. I had no real issues with the samples before (you'd basically only notice significant JPEG if the original image itself has been unreasonably compressed), but now I can see it even with basically perfect sources. Has nothing really changed?

1