Nuke resolution metatags

BUR #7423 has been rejected.

nuke lowres
nuke highres
nuke absurdres
nuke incredibly_absurdres
nuke tall_image
nuke wide_image

What is "highres" changes over the years and is purely subjective. Phone screens can be higher resolution than the danbooru definition of "highres".

These tags have been superseded by mpixels:2.., height:10000.., ratio:1:9 and so forth.

For legacy purposes, optionally, I propose to suggest and automatically place (for instance) mpixels:2.. when typing highres in the search box. This way, searchers become aware of these search capabilities.

These tags are hardcoded, thus require code update prior to this BUR being possible. The proposed nukes are expected to remove 2.9 million tags, of which the majority are highres (2.2 million) and absurdres (0.6 million).

What about wide image, should that be included in the BUR?

redtails said:

These tags have been superseded by mpixels:2.., height:10000.., ratio:1:9 and so forth.

Aren’t mpixels and ratio extremely slow and prone to timing out because they’re not indexed? I think the data isn’t even stored in the database and is instead calculated on the fly from width and height, making those searches even slower than other non-indexed searches. Nuking the resolution tags will make it much harder to actually find images of a particular (minimum) size.

kittey said:

Aren’t mpixels and ratio extremely slow and prone to timing out because they’re not indexed? I think the data isn’t even stored in the database and is instead calculated on the fly from width and height, making those searches even slower than other non-indexed searches. Nuking the resolution tags will make it much harder to actually find images of a particular (minimum) size.

mpixels and height/width are indexed. I don't think ratio is.

This isn't to say ratio couldn't be stored in the database in the future, of course.

Also worth noting is that the *res tags say nothing of ratio so regardless of ratio being a problem they can be replaced with mpixel/width/height searches.

lowres and highres are far more intuitive for the average user than something like 'mpixels'... which you would only even know existed if you read through the search cheatsheet.

So was wallpaper, but we deleted that and linked to relevant searches in the wiki.

Talulah said:

So was wallpaper, but we deleted that and linked to relevant searches in the wiki.

Because wallpaper was meaningless (as there are masses of different ratios that can be used as wallpapers so you had to use the ratio tags to actually get what you wanted already), controversial as to what it even should mean and barely used. Lowres and highres have single fixed definitions and are used on 100% of applicable images.

They have fixed definitions because we gave them fixed definitions. We could have done the same with wallpaper, but we opted not to.

Talulah said:

They have fixed definitions because we gave them fixed definitions. We could have done the same with wallpaper, but we opted not to.

We really couldn't. Different devices have different proportions. The only way to do it would be to have a 16:9 wallpaper tag, another 9:16 wallpaper tag, another 4:3 wallpaper tag, etc.

And if we'd done that then there'd have been no point in an overarching wallpaper tag.

Different devices have different resolutions too. How is that different? What Danbooru defines as highres won't even fill a standard 1080p display. These are just arbitrary numbers we've decided were appropriate, and they did not age well.

A highres image is still high resolution relative to a non-highres image regardless of your display settings.

Whereas a wallpaper for one screen ratio is completely useless as a wallpaper for any other screen ratio.

That isn't even necessarily true because of how the rules for tagging it are defined. Is post #119020 (453600 pixels, 160x2835) really higher resolution than post #2972175 (7669603 pixels, 3197x2399)? Despite being over 10x the megapixels, the latter is not tagged absurdres while the former is. This is obviously the most extreme example, but it just shows one more reason why these tags are rubbish.

skylightcrystal said:
lowres and highres are far more intuitive for the average user than something like 'mpixels'... which you would only even know existed if you read through the search cheatsheet.

I agree. Hence the suggestion to let the UI translate highres to mpixels:2.., therewith making highres unnecessary as a tag.

kittey said:
Aren’t mpixels and ratio extremely slow and prone to timing out because they’re not indexed? I think the data isn’t even stored in the database and is instead calculated on the fly from width and height, making those searches even slower than other non-indexed searches. Nuking the resolution tags will make it much harder to actually find images of a particular (minimum) size.

highres et al. represent roughly 2% of all tags. Nuking them speeds up literally every other search, lowering the overall burden on db. Ratio can be made a new db field and therewith should be fast. An admin will have to step in to illustrate how often highres gets searched for, but I expect that it does not weigh up to the expected speed gains of other searches.

skylightcrystal said:
A highres image is still high resolution relative to a non-highres image regardless of your display settings.
Whereas a wallpaper for one screen ratio is completely useless as a wallpaper for any other screen ratio.

highres is ~2 megapixels and up. I do not consider 2 mpixels "highres". A 1440p display is ~3.6mpixels, 4k display is ~8 mpixels. To me, illustrations with and without highres neither are highres. We already have mpixels:2.., so each searcher can define what they finds highres.

Here's the top resolution searches in the last 7 days. highres has around ~18k total searches, absurdres ~9k.

I've been wanting to get rid of the highres tag for a while. The problem is that while I think the current tag is useless, it does get a decent amount of search traffic, so I'm reluctant to nuke it without giving people an easy alternative. Aliasing it to mpixels:>2 or something else is problematic because aliases are meant to be for real tags, not metatags, and allowing metatags in aliases would cause problems in various places.

The other tags here like lowres, tall image, and wide image are more tolerable because they're less common. Highres is the main problem because nearly 75% of all new uploads are considered highres, which makes the tag useless and is pure tag bloat.

Instead of making it a real alias, for legacy purposes it could be a soft-alias, i.e. the search box suggests and auto-replaces mpixels:>2 when highres is typed. This way, searchers are kindly introduced to more advanced search strategies.

Further information:
mpixels:>2 -highres gives 0 results.

A significant amount of the search traffic for highres is likely from being easily discoverable (and thus clicked on) in the tag lists (as opposed to being entered manually in the search box).

This discoverability would not be preserved by the proposed alias (whether real or via autocomplete).

If you care about keeping resolution searches discoverable from the sidebar, there might be some way of doing that while nuking the tags, but not sure what that could look like.

Just for perspective:
Google image search considers images of around 0.8 MP or more to be "large".
Bing includes images as low as 0.4 MP under "large" and around 0.9 MP or more are "extra large".
Baidu considers images of around 0.7 MP or more to be "large" and around 3 MP or more to be "extra large".
Yandex also considers around 0.8 MP or more to be "large".
Considering this, I think our highres tag returns results in line with what most users would expect.

These are estimates from a few searches and using a script to find the smallest result.

Nameless_Contributor said:

Just for perspective:
Google image search considers images of around 0.8 MP or more to be "large".
Bing includes images as low as 0.4 MP under "large" and around 0.9 MP or more are "extra large".
Baidu considers images of around 0.7 MP or more to be "large" and around 3 MP or more to be "extra large".
Yandex also considers around 0.8 MP or more to be "large".
Considering this, I think our highres tag returns results in line with what most users would expect.

These are estimates from a few searches and using a script to find the smallest result.

For comparison, the baseline 1600x1200 image is 1.92MP, the minimum possible size (1x1200) is 0.0012MP, and the largest possible size (3199x2399, above which is absurdres) is 7.67MP. The average highres image is 1.93MP (slightly above baseline), which has been increasing over the years:

So, if you consider 2-3x to be "in line", then I suppose we are.

evazion said:

I've been wanting to get rid of the highres tag for a while. The problem is that while I think the current tag is useless, it does get a decent amount of search traffic, so I'm reluctant to nuke it without giving people an easy alternative. Aliasing it to mpixels:>2 or something else is problematic because aliases are meant to be for real tags, not metatags, and allowing metatags in aliases would cause problems in various places.

The other tags here like lowres, tall image, and wide image are more tolerable because they're less common. Highres is the main problem because nearly 75% of all new uploads are considered highres, which makes the tag useless and is pure tag bloat.

But that's just a problem of any big tag if we are being honest.
At a point in time, tags lose their specifity while still being usable per se; in other words, people use this tag because they have a good understanding of what to expect even if an "long image" appears in the search every 50th page or so.