Winter sunset silhouette at the Lorain West Breakwater Light in Ohio on Lake Erie. Photo by Tony Psaltis.
#LighthousesOfMastodon
#GreatLakes

Tom Hanks movies can go either way — there are ones that I’ve absolutely loved, and then there are ones that I care nothing whatsoever for (and I’m looking at you, “Finch” …)

But “A Man Called Otto”, what can I say? It feels like his best movie ever, at least right now just after watching it, feels like one of his very best 🙂 I loved everything about the movie and how the story unfolded. It was a brilliant performance, and a great story!

#MiniReview #Movies

Mount Fitz Roy
Patagonia, Argentina
#nature #geology #wonders #mountain #summit #mastadon

Ocean Spray #art #mastoart #watercolor #seascape

@BobWilliams Sure thing 🙂

I just like finding solutions and I do like knowing that a JS solution is there if I want it. So any poking around is not (just) for your sake 😛

There were a couple of more advanced JS clients that I took a look at but I think they either got too complicated coding wise or didn’t have the features I wanted. I like to keep things as simple as possible …

But in case it helps, here are the links to those too:

https://github.com/oken1/kurotodon

https://github.com/nolanlawson/pinafore

https://github.com/NickColley/semaphore

@BobWilliams If I don’t do something immediately, I (generally) forget to do it later … So I did a quick GitHub repo, added a README and updated the code. Here it is:

https://github.com/FahimF/FediJS

It used to run fine a few months back but apparently has issues now 😛 I did a few quick fixes to get it to show the login screen but can’t get beyond that at the moment since I originally did the development for a Mastodon instance and I’m now on Akkoma and Akkoma recently changed their auth mechanism.

I don’t know for a fact that that’s the issue but hopefully, if you are on Mastodon, it just works for you. Will take a look perhaps this weekend to see what the issue is but would be happy to discuss further if you run into issues too …

And that’s it for papers today — 6 papers boosted from the cs.CV category (and one outside) out of a total of 98 on arXiv.org today ...

#AI #CV #NewPapers #DeepLearning #MachineLearning

"Accuracy and Fidelity Comparison of Luna and DALL-E 2 Diffusion-Based Image Generation Systems. (arXiv:2301.01914v2 [cs.CV] UPDATED)" — Comparing the accuracy and fidelity of images generated by DALL-E 2 and Luna, which is Stable Diffusion-based.

Paper: http://arxiv.org/abs/2301.01914
Luna code: https://github.com/slowy07/luna

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Selected image samples from the…

"An efficient deep neural network to find small objects in large 3D images. (arXiv:2210.08645v2 [cs.CV] UPDATED)" — What it says in the title 🙂

Paper: http://arxiv.org/abs/2210.08645
Code: https://github.com/nyukat/3d_gmic

#AI #CV #NewPaper #DeepLearning #MachineLearning

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

"Leveraging Large Language Model and Story-Based Gamification in Intelligent Tutoring System to Scaffold Introductory Programming Courses: A Design-Based Research Study" — Using Large Language Models (LLM) and gamification to teach programming in a more digestible format (especially) in introductory programming courses.

Paper: https://arxiv.org/abs/2302.12834

#AI #NewPaper #DeepLearning #MachineLearning #HumanComputerInteraction

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

"Diffusion Posterior Sampling for General Noisy Inverse Problems. (arXiv:2209.14687v3 [stat.ML] UPDATED)" — Extending diffusion solvers to efficiently handle general noisy (non)linear inverse problems via approximation of the posterior sampling.

Paper: http://arxiv.org/abs/2209.14687
Code: https://github.com/dps2022/diffusion-posterior-sampling

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Solving noisy linear, and nonli…

@BobWilliams Sure 🙂 Will put up the JavaScript code on GitHub at some point today and let you know … It’s very basic at the moment but it will at least let you connect to an instance and see your home timeline. From there to filtering should be fairly straightforward since all you need to do is go through the fetched posts and figure out what you want to filter out … (hopefully) 😛

@BobWilliams If you are on macOS, I’d be happy to post binaries in one of my GitHub repos so that you can try out the functionality now …

Otherwise, you’d have to wait till I get around to doing the JavaScript client 🙂 I don’t think I posted the source for it online since it’s way too early in development but if you’d like to play around with that, I can post the source for that on GitHub too …

"Subspace Diffusion Generative Models. (arXiv:2205.01490v2 [cs.LG] UPDATED)" — Restricting diffusion via projections onto subspaces to reduce computational time and cost without affecting the overall quality of the generated image.

Paper: http://arxiv.org/abs/2205.01490
Code: https://github.com/bjing2016/subspace-diffusion

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Random high resolution samples …

"Large Scale Visual Food Recognition. (arXiv:2103.16107v3 [cs.CV] UPDATED)" — A food dataset with 2,000 categories and over 1 million images that can be used for food recognition.

Paper: http://arxiv.org/abs/2103.16107
Code: https://github.com/Liuyuxinict/prenet/

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The distributions over each cat…

@BobWilliams I guess it might be possible to do something from the browser end but for that I’ve only explored things like writing a browser-based client in JavaScript 🙂

That’s probably the solution which would cater to most people since it should run on most (any?) platforms but it turned out that there wasn’t an existing pure JavaScript solution that I could leverage. I’d have to write all of it on my own. I got started but didn’t go beyond displaying a timeline since all the work that remained seemed too daunting and it was too long to wait for the payoff.

So I just found the best native (macOS) client that was open source, and modified it to add the features I wanted. But I’ll probably go back to the JavaScript client at some point since it should help a lot more people (possibly) …

@BobWilliams Making changes at the instance end would be one way to set things up for a larger group of people I guess 🙂

For me, personally, there aren’t a lot of Fediverse implementations which have the features I want and the ones that do are implemented in programming languages that I’d have to learn, or the implementation itself is rather cumbersome, as is the case with Mastodon.

So I opted to do the changes at the client-end with specific client apps since that’s a little bit more manageable. That presents certain problems but less so than managing the server-side stuff to do things in an efficient way … at least for me.

Plus, if you get the client working right, you can use it to connect to any instance and so be able to move to a new server if you wanted, etc. 🙂

"Directed Diffusion: Direct Control of Object Placement through Attention Guidance. (arXiv:2302.13153v1 [cs.CV])" — Controlling object placement in diffusion models by way of attention guidance.

Paper: http://arxiv.org/abs/2302.13153

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Directed Diffusion (DD) key res…

@BobWilliams Happy to help in any way I can 🙂 I’ve been running some experiments myself in reducing my home timeline by filtering but only by way of customising my client app since the generally available apps don’t seem to have that feature.

I find lists much more useful now that I can filter out list members from my home timeline but somebody else said something which leads me to believe that other people might be disregarding the home timeline altogether (since they can’t filter by lists membership) and simply putting all of the people they are following into lists so that they only deal with lists.

Not sure if that’ll help you but just mentioning in case it does …

A total of 98 papers in the cs.CV category on arXiv.org today — 35 new, 63 updated.

Gentlepeople, start your engines 🙂

#AI #CV #NewPapers #DeepLearning #MachineLearning