"Which country is this picture from? New data and methods for DNN-based country recognition. (arXiv:2209.02429v2 [cs.CV] UPDATED)" — A framework to identify the country where an image was taken, which could be useful in debunking fake news and many other applications.

Paper: http://arxiv.org/abs/2209.02429

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Three sample photos taken from …

"Consistent Diffusion Models: Mitigating Sampling Drift by Learning to be Consistent. (arXiv:2302.09057v1 [cs.LG])" — A possible solution to fix the sampling iterations drifting away from the training distribution when generating images using diffusion models.

Paper: http://arxiv.org/abs/2302.09057
Code: https://github.com/giannisdaras/cdm

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Visual comparison of EDM model …

"LayoutDiffuse: Adapting Foundational Diffusion Models for Layout-to-Image Generation. (arXiv:2302.08908v1 [cs.CV])" — A method for generating images based on a semantic layout which describes the positioning of the various component elements in the final image.

Paper: http://arxiv.org/abs/2302.08908

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
LayoutDiffuse is able to genera…

@broximar Back then, I think I would have played anything since there wasn’t much to do beyond read 😛 I knew this guy who had a huge collection of games and he would grudgingly give you a copy of a game if you gave him something new.

Given that there was no Interenet (and BBSes were not quite heard of in Sri Lanka at that time), this “sneakernet” was the only way to get games back then where I was. And everybody wanted the popular stuff like “DigDug” or “Need For Speed” — not “Sim” games 😀 But yeah, “SimLife” was disappointing after all the stuff I’d conjured up in my head based on the description …

@mairhart

"Paint it Black: Generating paintings from text descriptions. (arXiv:2302.08808v1 [cs.CV])" — Explores generating paintings based on text input.

Paper: http://arxiv.org/abs/2302.08808

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Examples of paintings generated…

Talking about “Sim” games has brought up a lot of old memories 🙂

@mairhart reminded me about “SimTower” — I still remember the cover that a British gaming mag ran for it, but I don’t remember the name of the magazine 😛 It was a rather interesting entrant in the “Sim” series and got you to do a lot of the city management stuff on a more smaller (but not quite microscopic vs. Macroscopic) level.

Then there was “SimAnt” which I think was all about managing an ant colony? I know I played it but I remember very little about it now …

“SimLife” was what I was really excited about back then because I thought I could create new electronic life, run my own planet, etc. But in reality, if I remember now (30 years later) it was rather boring 😛

But there’s also all the “Civilization” spin-offs. I knew I was forgetting something, but for the life of me I couldn’t remember “Alpha Centauri” till I looked it up 🙂 I still don’t remember much.

And speaking of forgotten games, apparently there was also “Sid Meier's Starships” — despite being one of the newest in the series, I remember nothing about this either. I don’t know if I knew of it and forgot or if I’d just lost all interest in “Civ” games by then and simply didn’t care. But I think I’ll try to find a copy and play it just to see how it goes …

#Gaming #Civ #Sim #Reminiscences

"Fine-grained Cross-modal Fusion based Refinement for Text-to-Image Synthesis. (arXiv:2302.08706v1 [cs.CV])" — Another text-to-image generation approach where, instead of generating the final image from a noisy image, you generate an initial low-resolution image based on the input text and then use a GAN (Generative Adversarial Network) during the second stage to generate the final output.

Paper: http://arxiv.org/abs/2302.08706
Code: https://github.com/haoranhfut/FF-GAN

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Samples generated by Attn-GAN, …

"Foundation Models for Natural Language Processing -- Pre-trained Language Models Integrating Media. (arXiv:2302.08575v1 [cs.CL])" — An open access book about Natural Language Processing (NLP) models and how to build and improve them.

Paper: http://arxiv.org/abs/2302.08575

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>

A total of 94 papers in the cs.CV category on arXiv.org today — 55 new, 39 updated.

I just want to work on my list filtering code for my Fedi client, but I guess I’ll have to go through papers first? 😛

#AI #CV #NewPapers #DeepLearning #MachineLearning

@mairhart Yes!!! The old-school gameplay, while it might not have had the richness of the graphics, had such a lot of fun elements. I loved playing games over and over. Now, it just feels like work even when I think about it … You have to wait for things to load, monitor resources (computer not game) and it all seems so daunting.

“Beyond Earth” felt so much more fun but then they really didn’t seem to want to concentrate on that. I really think with a great “alien exploration” story it would have done so much better. Ah, well ….

@mairhart Yep, “Civ V” was where they lost me too. It looked great but the gameplay just took too much effort, especially on macOS where it would heat up the machine like crazy … I just couldn’t get into it and (possibly) having to shutdown every once in a while when the machine got hot didn’t help. I don’t think I’ve recovered from that …

I had forgotten about “SimTower” … and “SimAnt” 😛 There are a few others I’m forgetting too, but I agree, “SimTower” was a great new approach to the whole franchise and they probably could have done a lot with it but back then, “Maxis” was churning out Sim games like crazy and I don’t think they were interested unless a game was doing really well …

@at I wouldn’t say “energy” exactly … but I’m very picky about what I want and if I can’t have it, then I guess I’ve got to create it 😛

So … good news, bad news situation with yesterday’s experiment with putting some of the more high-volume people I follow on to lists and unfollowing them 😛

My hometimeline was so much leaner and cleaner today. There were very few replies and I just got posts that I was actually interested in and was through my home timeline in record time since I had like one third the normal volume of posts.

On the other hand, the list had not moved at all since I created it yesterday, except for one post 😀

Now if I’d stopped to think about this before I came up with my “brilliant” plan, I would have realized this — given the philosophy behind Mastodon/Fediverse, I don’t think anybody really wants you following people without the followee knowing/approving the action. So, creating lists the way Twitter allowed you to would not make much sense from a privacy standpoint.

Unfortunately, that makes lists a rather useless feature in the Fediverse since, as far as I know, there is no server-side filtering of posts from lists from your main timeline. If you just scroll through and read a bit and don’t care about duplicates, this is fine. But if you obsessively go through every post till you are caught up, like I do, then this becomes annoying pretty fast 😛

So my plan at the moment is to implement client-side filtering. When posts are fetched from the server for the home timeline, see if any of the fetched posts are authored by, replies to, or boosts by people on any of my lists. If they are, filter out those posts since they will be displayed on the lists.

I’m not sure how CPU intensive this will be and how smooth the experience will be. But I’m going to try and find out …

#Lists #Following #FediStuff #FediClients

Just a cottage by the water. Sometimes simple scenes are best. #aiart #midjourney

Some years ago the news that a new “Civilization” game was in development would have excited me to no end 🙂 I used to love “SimCity” and “Civilization” to no end and I think back in the day, I played every “Sim” game that I could get my had on … “SimEarth”, “SimLife”, “SimFarm”, “SimTown”, “SimCopter” … I think I played all of them … or at least got them and tried them …

But, I didn’t play “Civilization VI” and the new game brings me no excitement. I think I just can’t get excited by that kind of gameplay any longer. Sure, compared to the original “Civilization”, the graphics have come a long way and it looks gorgeous, but I just don’t have the time/energy for the game play.

Now if they come up with a new version of “Civilization: Beyond Earth” with a good storyline (not just the building and exploration), I might be up for that 🙂

#Games #Civilization
A new Civilization game is in d…

@at To update you on the lists situation, well, the pruning did work in giving me a nice manageable hometimeline. There’s only like one third the notifications from overnight (of course, this is a weekend so the volume is lower) and I am able to make my way through the reduced timeline so much easier.

But … you were right about the lists 🙂 Akkoma appears to work the same as Mastodon in that you have to be following someone in order for their posts to show up in a list.

I did look this up for Mastodon and there’s an open feature request asking that either users be able to have lists without following people or be able to filter out people in lists from their home timeline but there doesn’t seem to be much movement on it.

So I think I’ll have to end up doing this client-side. Just filter out posts/boosts on the home timeline from people who are on my lists. This should be easy enough if you have source code to the client and that should theoretically allow you to have the functionality I wanted. I guess I’ll try that and see …

#silentsunday

In China there is a legend that if a carp is strong enough to swim upstream against strong currents, it can become a dragon.
#SwampSunday #ukiyoe #folklore #JapaneseArt @folklore
🎨'Carp' from an untitled series known as 'Large Fish' - Utagawa Hiroshige, 1842.

@qb As far as I know, there is no max posts per day setting 🙂 I did know about the setting from your screenshot but always assumed that that was only for the web frontend and I use a third-party client. But I should have verified …

Also, it’s possible to add some of the stuff that I want to third-party clients and I might actually have to end up doing it that way since somebody else mentioned that lists might not work for what I had in mind (at least on the Mastodon side) since lists only show posts if you follow an account ... I guess I’ll see after testing for a day or so …

@at I’ll know tomorrow morning since early morning is when the enormity of the timeline hits me 😛

I don’t know if Akkoma works in a different way with regards to lists since the list I set up is populated, but I set up the list before I unfollowed each of those people. So again I guess I’d have to wait a day and see if the list gets any new content or if it is only the old content from before I unfollowed.

I guess based on the Masto/Fedi philosophy of putting privacy first, having lists not be populated if you don’t follow the person makes sense. But the trouble is that if you mute a person then they are removed from both home timeline and lists. So that doesn’t help either.

I just want a way to have a manageable home timeline and then possibly a set of lists where I can put other interesting but high-volume people. Will let you know if I have any success 🙂