/hydrus/ - Hydrus Network

Hey hydev, I have a question. Could it be possible for you to make some way to convert fullwidth english characters to normal? I can't type them, so I can't search for pixiv titles that have been imported which use them. Also, could there be some way to get around normally untypable characters like "・"? Maybe some configurable table to change characters.

Every time I'm getting my dupes close to being under 30k, I download a bunch more files and the dupes go up to like 32k again once I recheck.

I've realized this is largely because Tumblrites may keep good image blogs sometimes, but they also constantly feel compelled to reshare compressed versions of images and to put shitty filters on everything.

Dev, please patch Tumblr poster thought processes so this stops happening, it's quite disruptive to my workflow.

>>9440

lol wish I had this problem

I got 100k right now, and if duplicate worked for my whole db, I would have around 180k more at minimum. honestly wish I had only 32k images to go though

>>9429

Hi dev, could you make hydrus use wxversion? I have wxPython 3 & 4 installed and by default 3 is used, which won't let hydrus start. I modified the first couple lines of ClientController.py and it seems to have fixed it.

import os
import wxversion
wxversion.select('4.0.1')
import wx

>>9441

Yeah, I forgot about Hydrus and imageboards in general for a year or so due to my real life getting busy. I still don't have time to monitor or collect everything I'd like to. It'll be better when the download and login engine are both done and working together, though.

Does Hydrus have a slideshow functionality in its viewer? I'd like to have it open on one monitor and changing say every 30 seconds so I can pick out my favorite images or do a little light tagging as it goes.

I'm thinking either full random drawn from entire database, or from inbox, or just a sequential auto slideshow through some search results for a tab and I'd just search limit 1000.

>>9447

Yes, you can do a slideshow by going into the media viewer and then right click > start slideshow and then picking an interval.

>>9448

Thanks man, that's perfect.

>>9453

>jpegs and pngs

FLIF support when?

>>9453

I honestly doubt hydrus is the cause, I think it's controller related again, and hopefully I finally got that shit solved.

On the dup side, I have an idea. You have a 'this is better' option, you could split off the duplicate there.

lets say

you have a new duplicate

it goes into the normal 'here is duplicates' sort

You get another duplicate that is tied to a duplicate that already went through a 'this is better' sort. now you the duplicate go into a seperate sort, this potentially makes it VERY easy to mass sort images. If you have a larger file, or a larger file size, you could quick kill all duplicates there, but if the new file is bigger either in size or file size, it could be left for you to manually sort in a duplicate 'contender' filter

I think that could take much of the pain in the dick out of a massive duplicate to do list.

I also think that having duplicate filter opening both images at once side by side would also make it massively easier, mirroring moves and zooms between the two but I have no idea how viable the option is, really any massive rework for this seems like it would need massive work in other areas too.

that said, ran into an issue where a multi thread watcher had a thread that 404'ed and it refused to stop checking it, so there was a perpetual 172/185 till I killed it.

>>9453

>Hey, I am sympathetic to your problem, but I think I read that we aren't supposed to use wxversion any more with wxPython?

Ah, I was not aware of that. I'll try venv, thanks for info.

>>9448

But machine learning can't generate the exact image again if you lose it after forming an irrational chemical preference for it, so the ride must continue regardless.

In fact the ride will just get longer after ML programs start spamming out near-duped generic waifus that everyone here will love anyways.

Really though, I don't mind dupe sorting, I just want an option that lets me scroll through all detected dupes of a file at once in fullscreen, and after a lot of thought, I think I just now figured out how you could do it simply and in a way that doesn't need many lines of code and doesn't confuse users much, so here it is:

>add a checkbox to dupe handling options "allow multiple dupes in filter"

>add field below it which is enabled if you check the box "max dupes to show" for the number of files you scroll through, minus the first file you're dupe-checking off of - by default it's 1, which is just the two-image process it uses now

>this way if you have 1000 dupes of one image you won't have to scroll through each one checking the image and the hints up top before you make a choice, you can just say max = 5 and it will only feed you 1+5 images to look through. 0 in the box means unlimited # of dupes shown and should only be done by people with small DBs and big beautiful RAMs

>options are exactly the same simple options as for the two-image comparison, nothing fancy, just choose one out of the group as better and delete/deprecate all of the others, choose alternates to mark all of them as alternates of the original image and each other (okay that one is more complicated but still, handle them as an array and it's just a nested loop), not dupes and same quality same deal, custom action could be a mess to implement but you could just delay it and remove that button from multi-dupe processing for now BECAUSE…

>in place of the custom action button you put a "swap to single-dupe" button which removes all files other than the original and the dupe you're currently displaying (or the first dupe in line if you're currently displaying the original image) from the filter and from the array the code behind it is using, so now you're back to single-dupe mode and your choice will only affect those two files. Good workaround for custom actions and also good for when someone needs to do something more carefully, so they don't have to leave the filter, go uncheck the settings box, and suffer with single dupe selection while they hope and pray that the right dupes come up for the thing they wanted to do

>when the filter feeds you a multi-dupe case, it displays how many dupes (or how many files total which is just dupes+1 if that seems less confusing) are listed in this choice, so the user won't forget to look at them all

>also the background colors change to something else, say shades of green, instead of shades of gray so you immediately know you're not doing a two-image choice, and this will just become an instinctual memory because color is simple

>the display of number of dupes is an always-visible small box, maybe in the upper-right hand corner as part of the box that shows if file is in inbox/archive/trash, or maybe floating just above it

>maybe that floating box also has a "switch to single-dupe" button in it, or maybe that's the only location of the switch to single-dupe button, so it's always visible and available

>maybe both single-dupe and multi-dupe (with limited # shown) have a toggle option to process through all dupes of one image before going on to the next base image in the filter

>that's it, you're done and you're mostly just reapplying single-dupe sorting functionality, but it's way faster if you have four or five of the same image and can just scroll through and pick one

I mean, it still won't auto-detect artifacting to save us from squinting, but looking at filesize is pretty helpful for that unless the file server is full on retarded and re-bloats artifacted stuff

>>9455

FLIFBOIS UNITE

AUTO-CONVERT TO FLIF ON IMPORT/DOWNLOAD OPTION WHEN

(idk maybe when the codec is actually finished?)

>>9461

I do that, but instead of a billion tabs I just finish the qa/tagging/whatever I'm doing for one batch, then add a "done" tag or something and redo the search. I never try to completely tag or qa in one go, I'm just doing rough sorting for now. After all some of them are probably dupes so I don't want to dig in too deep and waste time yet.

Also lately I've just been doing 50 instead of 1000 at a time because I make less errors when I take a break to redo the search than just paging through for hours.

>>9466

with flif I thought of that too, I don't share images, so if my entire image archive converted to flif I would have very few problems, however the only ones I would want auto converted are png/other lossless images into another lossless codec, that way converting them out of it for share is lossless too, for jpegs, its image specific, but hydrus intrunal, as in the thumbnails, i'm ok with every single image being converted there and those are not for sharing at all, that and I have around 50gb of a ssd filled with hydrus thumbs. even if its an incrementally smaller size, it adds up.

as for me processing them, its honestly just a rough pass only rating things I want to keep track of as 'favorite' and things I need to keep track of 'favorite' and 'tag later'

till I can burn through duplicates, I don't really want to go through all my files and really get down to business with rating

that said, im rateing in gated ways

pony/not pony

funny/not funny

art/not art

safe/suggestive/explicit

real/drawn

tag later/nothing

favorite/nothing

for the most part this massively culls images into areas that I can whittle down what I want to deal with.

Art is largely is this an artistic pose/something of value to learn from rather then something line anime

read drawn segregates images fairly hard, but the first hard gate is safe/suggestive/explicit denouncing not porn, I don't consider it porn, porn

If you are dealt with with gate 1, you will be rated that no matter what, while most other things are modifiers, real/drawn is also a gate status, however honestly getting that and is it porn can be done in one go, createing 6 distinct categories for images, ill likely gate off the drawn with eastern or western style, then inside each style, comic page or not, from there stand alone or multi image, with this being an end state as there is no way in fuck im tagging every god damn comic page, however I would also set a 'tag further' status to the ones that are worth tagging.

I like working in ratings more than tags as its far less of a hassle to set them then it is a tag currently.

What I want is something that can be expedient, because if it took me a minute to tag an image, and in said minute 6 more images came into the archive, it would take a depressingly long time, however if it takes me 1 second to see the image, determine what it is drawn or real, another second or two to see if its porn or not, a once over if its funny, I could possibly rate at a pace faster then my import, however before I even consider doing that, I have to have a way to notate an image and delete it, then see the notation on why it was deleted, otherwise I am needlessly triggering ocd adding any amount of images to the archive. as now I have to check 'why was this deleted, was it a mistake' and 2.3 million files of which I may only want to keep 1 million, along with around 280k of them being duplicates or potential duplicates… my image archive is a fucking nightmare., thank christ for hdds getting bigger faster then I use space.

Just found out about this program.

I was totally expecting to see a dead board but it's nice to see that this is getting updates

though i'm not sure why you would need weekly updates for this kind of thing

>>9453

>Some additional error information is supposed to pop up if you try to open externally if you have help->debug->report modes->callto report mode turned on

I get this

>2018/07/24 22:36:39: Attempted to launch /mnt/ext4/Hydrus/db/client_files/f51/51a1fb14b8227f0f27b1d6d1278deefa2f3cdbab0b7977cf6eab6a38c539dd27.webm using command ['mpv', '/mnt/ext4/Hydrus/db/client_files/f51/51a1fb14b8227f0f27b1d6d1278deefa2f3cdbab0b7977cf6eab6a38c539dd27.webm'].

>2018/07/24 22:36:39: No stdout or stderr came back.

>>9473

Well, for now I've replaced the liblzma file in the hydrus directory with the one I had in /usr/lib/ and it works. If fixing it is a huge mess it's not a huge priority for me as everything seems to work fine like this.

>>9471

weekly updates… go back a few versions and see what changes week to week, some changes are massive, some are incremental

take a look at when multi thread watcher got implemented and each release after improved it. the most basic form of it was workable and a huge resource saver, in fact I don't think I can possibly make the program crash in the same ways I use to since it got added

I think I have somewhere around 500-700k images in in thread watchers just waiting to be used. dealt with, a good number of these are already in archive and are duplicates so I'm holding off the mass processing, but having them all separated into themes that can be quickly parsed and feel like I did something… that's really nice.

When importing files with 315 the order is always forced to be by name, this is quite disastrous for me as i always sort my import folder by date created and then drag them into hydrus to save that time (since hydrus only cares about imported time and not created on pc)

I mostly download images from boorus with the booru tag parser, and that gives them an hash filename, being forced to having them being uploaded by name order is messy enough to keep me from using 315 (the "solution" would be to use a renamer to have the name reflect the date created time, but it's a mess because of the paired txt files for tags)

Anyway, is this an intentional change or a bug? Is there an option to get back the drag&drop order? I really can't use 315 because of this

>>9466

>>9455

The only real impediment to me adding more image format support is library support. I use Pillow (PIL) and OpenCV to do all my image rendering, so if either of them add FLIF or HEIF, I can usually add support in hydrus within a week. Or, if you can point me to a simple library on PyPi (pip) that can take a FLIF and give back simple metadata like resolution and go path->bmp no prob, please do and I will check it out.

My general feeling on this is that these are all meme formats for now, and everyone is waiting to see which is actually going to be useful IRL for real problems. WebP is another. I am personally hoping for high colour depth (HDR, proper greens etc…) to take off in the next five years or so, so I am personally cheering for >8bits per channel. A quick look suggests FLIF does up to 16, so it gets my vote. iirc, WebP is stuck at 8. I haven't touched HEIF much, but I think it has a gif-tier mickey-mouse container format that supports some kind of animation–my rough guess is it is Apple-backed corporate bullshit that may only actually work well on iPhones.

Is there a good lossy format in the pipeline? Something with high colour depth, decent compression and CPU/GPU usage, and no over-engineered 'it streams great from our legacy CDNs/proprietary hardware' complexity?

I'd ideally like a format that could do both lossless and lossy, and then we could shove the default decision on how to encode image data to the computer and cut the normies completely out of the loop. But maybe the companies would make the default '80% quality' shit, ha.

>>9456

Thanks. Some of this 'deal with larger groups of potential dupes when some relationships are already known more efficiently' logic is in there, but the whole duplicate system is awaiting a rewrite from the current 'masses of pairs' logic to 'defined groups', which will make figuring out which pairs to present and in which order a lot easier.

My very very old 'figure out rating' filter had side-by-side, and it was unfortunately a workflow and maintenance disaster, enough that I deleted the whole thing. Having the quick flick back-and-forth is much better for seeing differences, although again I'd like to improve how the viewer works at high zoom and so on and so on.

Do you happen to still have that 404-erroring URL? If it happens again, please check the URL and send it on. And visit it in your browser, just in case there is some unusual or broken 404 that is being generated or something.

>>9466

>>9461

>>9469

I like to do something like:

system:inbox

system:limit=512

rating:has rating 'space babes, very important check these first'

You can clear them out in batches without making your session sluggish, and just hit f5 to throw up another random sample of whatever the rating category is. I've got a bunch of named ratings that I have mapped on shortcuts, so when a thread is dead, I can just ctrl+a and then mass-give them an appropriate quick search rating so I can find that pool of the inbox again.

A side thing on FLIF and high colour depth etc…: I assume that if/when we move to HDR displays, old jpegs and anything else 8bit are going to look a bit flat, in the same way you can tell a gif by the shit palette. Any mass conversion to a new format we do will be tinted by this legacy issue. It isn't as big as going from black and white to colour, but I think it may matter, and maybe we'll end up having neural networks doing 'colourising' for deep greens or whatever, since that tech is happening at the same time.

>>9466

For advanced ML, my guess is that the AI are quickly going to be so good, we'll just fall in love with the new stuff (and them by proxy) and soon forget about our broken old human relationships, and then a few years later, they'll just kill us all in our sleep out of pity. So it probably isn't a long-term issue to think about. :^)

Thanks for your thoughts on the dupe stuff. I like this idea. The original dupe system works on collections of pairs, which is the biggest thing holding me back right now–I'll convert the db side to 'groups', which will be much easier to deal with, logically, for larger decisions like these. I've still got about a hundred quality-of-life jobs to do from the initial dupe work and I am buried in the downloader now, but maybe I can catch up on some of this stuff and fit in the db change (which will also affect siblings and parents, which have very similar issues) in the interim between the download job finishing and the next big job (chosen by user vote) starts.

>>9478

>>9471

I'm a leave-my-homework-until-the-night-before kind of person, so if I have a release period of a month, I won't do anything significant for the first two weeks at least. This weekly schedule has kept me ticking ok for a few years now, although I still have problems managing my overall todo and thinking in the more strategic 'where is this going' long-term. It is also helpful for quick iterating with feedback–I can try out a button for some odd, complicated task, and if it doesn't quite do the job for the users who care, I can change it again the next week.

Another nice thing is when I fuck up, I can roll out a fix in a few days without too much hassle.

If you get into hydrus, you don't have to update every week if that doesn't work for you. Plenty of users only dip in once every two or six months or whatever. Let me know if you run into any problems!

>>9473

>>9474

Thanks for this. Linux environment shit is all greek to me, but yeah, it looks like the process started by hydrus is inheriting the cwd or something pyinstaller set up and so pulling .so files in from hydrus dir instead of mpv. I will have another look at this, see if there is some flag I can set up to say 'new process instead of child process' or whatever works.

>>9495

Thank you for this report. This was due to the new 'sorting by human numbers', which sorts 'blah - 99.jpg' before 'blah - 100.jpg'.

I am likely going to be full up doing gallery stuff for the rest of this week, but I will put a checkbox for 'keep original sort' onto that dialog this or next week. Please hang on to your to-be-imported files for now and give it another go later (or roll back to v314, I am not sure how doable this would be).

>>9503

I think flif has a "lossy" mode through progressive decoding or whatever is called where you can just use the first however many bytes as a lossy version

Is it possible to make hydrus "forget" the duplicate relationship after the images get deleted, or at least have it as an option?

This is my issue : i filter some duplicates better\worse and all is fine, but then let's say months later i come across the same picture and i upload it again since i forgot about it

Now, because months earlier i already filtered that image as a worse counterpart than the one i have, the next time i launch the duplicate filter that pair is not gonna show up since for hydrus it was already resolved, leaving me with a pair of duplicates that'll be almost impossible to find

The only way to see it would be to reset ALL duplicate relationships, or check in system better\worse once in a while to see if you accidentally reuplodaded a duplicate

It'd like it better if hydrus presented me the pair again in the dupe filter, it'd be better than having seemingly invisible duplicates around

Also in the better\worse decision, i'd prefer for hydrus to _not_ strip the worse one of all its tags and copy them on the better one, if it ever got uploaded again (even if by mistake) it would be a tagless image, making it even worse to locate in case of duplicates, also sometimes you don't even want to transfer the tags over but hydrus will do it anyway (unless i'm missing an option to disable this), but it will not transfer the rating (would be nice if it did)

>>9515

You can change the way duplicates give tags and ratings, however when I last did a test it was broken/I didn't get it to work.

also, is that the way that hydrus works?

Just downloaded the zip and tested… wow, that is how it works.

I don't know about tags because I am not fucking adding the repository to a test install, but yea,

So h dev, personally I don't want want that they are duplicates to be forgotten, however, a button/option to go though known same pairs would be nice.

here, I go though dups, I find duplicates, I delete the low res one.

sometime down the road I re import it for some reason, lets say its a batch of files that I want all imported regardless of what it is.

Now both files are back in, a new filter to recheck would be good. hell a recheck for every option would be good, potentially tedious, but good.

lets say I have a bunch of alternate versions of an image, it would be nice to be able to go through the alternate pairs, and same quality again, if only to check your work.

>>9517

Could you point to where i could change the way the duplicate filter gives tags and ratings? I'm afraid i can't find where that option is

Anyway, of course i wouldn't want the duplicate to forget the duplicate relationship by default, but having it as an option would be handy

It's really just for the cases where you upload again images that were already filtered and deleted in the past, it's bound to happen at some point, but hydrus is not gonna show the pair again in the filter because it still remembers the relationship

But i really need to count on hydrus to tell me i'm reuploading worse duplicate images, since i'm bound to forget and make mistakes i need it to tell me when it happens by showing me the pair again in the filter (of course only if it concerns previously filtered&deleted images, i don't want to go through every single group of duplicates again every time i reupload a past dupe)

Add vim shortcuts and i'll suck your dick.

>>9518

In the filter scroll the mouse to the top, and at the center top there is a gear, that should open up a way to change what happens when you mark something as better.

Personally I have been rating things more so tagging so my interest came in with that, but I don't think I was able to get it to duplicate tags, and now that I know the way it works is if I test it and reintroduce the files to the archive, hydrus decides 'nah, ill never show you them again' I don't really want to test further.

I Also have a problem where my database fucked itself because of power outs and an ssd, recovering it worked, but as a side effect I have 2.3 million files in my archive, but only 1 million files are in my duplicate file processing tab. Its an issue hdev knows/knew about at the time, but I told him dup processing is a curiosity till various things get implemented, namely a way to notate why a file was deleted.

this could help you because if duplicate processing notates an image, that note survives being deleted, and if you accidently re add a dup to the archive, if duplicate filtering adds a note that says "Duplicate - Worse" to the note, it would be possible for hydrus to search it in the future. not going to lie, its a bit of nigga rigging, but for your purpous the note system and a searchable note would work. ALSO, on import you would have a log of the files, if user notes show up you could see you accidentally imported a previously confirmed duplicate image.

for me the system to show notes, and ability to quickly notate things on deletion expands functionality of hydrus so much as it allows me to delete files, give a reason, and move on because right now if I see something was deleted I have to see why. because of this, I think I have only fully deleted 300 maybe up to 1000 images, when I have a duplicate filter telling me I have 100k images to go though, and once notation is in and my dup process is fixed, i can add at least 180k more images to it, along with going though my images and culling them.

>>9522

tea thats what happens when you mass download images without culling though them

>>9524

if we are talking about quality, not the subjective is 'this my taste' but how good the image is, the one with gas fumes has little to no compression artifacts, the one on the left however looks like its been through a few passes of saving a jpeg as a jpeg and lowering the quality each time.

situations like this is why I would like a 'prefer alternate' button, if the 'prefer alternate' was in the note that I could see on a log page, I would know the exact reason it gone

right now we got

this is better

same quality

alternate

not duplicate

While I think we agree that the one without fumes is the better image, a strictly 'this is better' at least in terms of duplicates says they are both the same image, but an alternate image doesn't to anything with them, which is why a 'preferred alternate' button I think is the only option that missing. I think there are alot of people who have an image set thats 20 or so images, largely all the same image, just slight variance in facial expressions, this can/will/does create a duplicate image chain where they may all get caught they may not all get caught but realisticly you have 1 maybe 2 images you want to keep in that gallery

I imagine this is how the system would work.

lets say you get a file that has a chain of 20 or so images for the sake of argument and the one two images here >>9521 are part of it. you flag the left as the prefered alternate and it goes on saving that image for later. as there are still at least 18 more images it could be pitted against each other with

everything goes smooth, you finish, and you are left with 2 images,

1 is the prefered alternate

the other is a higher res/quality of the same image

but they never came up to each other in the general filter

Then you get another option, the prefered alternate filter

You then go through what's left there and see if what's left is a better worse pair, in which case you label it better worse. or the images are part of a set and you wish to keep both, so they are now alternates.

The main reason for this at least in my opinion would be to help with image sets.

The images that are prefered would be taken out of normal filtering and set aside for special filtering, so you don't constantly question 'is this the image I wanted?' and instead the image you want gets pulled out and set aside. after you go through all the duplicates, you may have something where its 2 or more prefered alturates or its 1 or more prefered and a contender image

it would then pit 2 of the images together to either get taken out in a 'this is better' filter, or in an 'alternate' filter.

I said it above too, where a contender image is just a way to expedite filtering, let's say you have an image that survived though a massive filtering, and a new image gets put in, in all likelihood it's not going to be better than what you have, so instead of going through a normal filter where 2 unknowns are pitted against each other, the contender filter pits the images against 1 that went through a 'this is better' and came out the winner against the new image, this also makes it a bit easier to filter as if you only filter condenser images, you could just set a rule that any file size smaller then it, or resolution lower, is just automatically filtered, and you are only presented with ones where it's honestly a question whether or not the new image is better.

granted, there would also need to be an option to disallow an image to be a contender, as there are some images I have that are the best quality I can find, but god knows someone increase their size or fucked with it just enough to be a wallpaper.

I think these two filters/modes would greatly increase the ability of duplicate filter be less of a pain in the ass to do.

>>9506

I didn't realize this thread was still alive.

Glad to hear you like the ideas for the duplicates, my vote is also to get the downloader/login engines both done before worrying about duplicates. I spend most of my active time working through duplicates, but I spend ALL of my passive time downloading and sometimes I run out of things that need to be downloaded because I can't easily pull imagesets and tags from as many sites yet.

>>9507

Can confirm, I almost always wait until a version with only minor changes drops to update because that way everything is basically guaranteed to have any significant bugs already fixed.

Why does the duplicate system find a bunch of dupes at search distance 4, then no more at distance 5, then a ton more at distance 6? It seems limited to numbers powers of two. Just wondering, it seems odd. Being able to increase the possible dupes at a lower number at once would be helpful, it wouldn't feel as overwhelming.

can we get a downloader option for derpibooru in a future update? the parsers and URL classes seem to be working (because URL downloads perfectly bring all the tags and full resolution picture), it just needs a downloader to allow for mass imports through tags

>>9531

hydrus doesn't up front tell you the duplicate distances anymore, so I cant remember exactly but it seems like the lower the number, the more of an exact match is needed to call it a dup as in less wiggle room.

The program already has a form of duplicate filter, where it filters out exact same images, so some distances may be effectively useless as the only way to get them would be lossless images where you change 1 pixel, and the only reason to do that is evading duplicate image filters on a chan, which more easily evaded by just saving it as a jpeg.

But the further you go away from an image, the more you bring in these are not duplicates at all.

>>9535

I am asking why a dupe search distance of 5 finds no more dupes than distance 4 but distance 6 finds tons of more. It would be useful if there was more granularity (is that the right word?) in the system and each increase in distance would give you a moderate number of new dupes instead.

>>9539

In my experience dupe search distances only work if it's a multiple of 2 (including zero).

>>9516

>>9515

>>9517

>>9518

>>9520

>>9521

>>9526

>>9535

Thanks for all this. I am not working on dupe stuff at the moment, but I have written these points down and will keep them in mind. I am overall happy with the first version of dupe stuff but agree that it could really do with a full pass.

>>9519

I don't know much about vim–what sort of shortcuts are you looking for?

>>9531

I was talking to someone else who had this issue a couple of weeks ago and couldn't reproduce it. When I do 4/5/6 on my test machine, I get more each time. The fundamental object that is being tested here is a 8x8 matrix representing some simplified vertical/horizontal waveforms in the image (it is a DCT), so I wonder perhaps if certain classes of image tend to be 'mirrored' in some way on the two axes represented in the matrix, but I am not sure.

I'd be interested if you discover anything new about this. The max potential number here would be 63 or 64 but most of the action is in the 0-8 range, with anything above, say, 12, being completely useless, so I may end up re-doing the shape-based dupe detection to be a 16x16 matrix (i.e. 256 bits) to give us more resolution here, although this would slow search significantly, so it might be better just to put my time into better workflows. The big bottleneck now seems to be human attention (it takes ages to go through pairs by hand), so I wonder if I should attack that first.

>>9532

Yeah, this is highly requested, so I'll roll it in before this whole overhaul is done. There's some stuff here for it already:

https://github.com/CuddleBear92/Hydrus-Presets-and-Scripts/tree/master/NEW%20Download%20System

but I haven't tested it myself and it isn't easy mode yet. You might just want to wait for me to add it naturally.

>>9532

>>9564

Shit, sorry, I read that wrong. Yeah, I'll add a gallery parser and 'searcher' for it when I have that system ready, so it'll just pop into the gallery selector button soon enough!