@ajyoung On Apple Silicon, the initial model load times are way longer for SPLIT_EINSUM compiled models than it is for ORIGINAL models. The EINSUM ones sometimes take about 2 minutes to load while the ORIGINAL ones load in about 10 -20 seconds at most.

Of course, some of this also depends on how you have the model loading, but once loaded, the initial image takes about 2 seconds longer to generate but then subsequent images at 10 - 20 steps are very fast. But if you go over 20 steps on the DPM-Solver++, it takes much longer. I think I had one on 50 steps which never completed and after about 3 - 5 minutes I just cancelled it …

So there are a bunch of factors at play and also depends on how you load the models. I find the Swift apps the easiest since you just load the model and then don’t unload the model till you quit the app 🙂

quick thread, retweets of which would be most welcome. just got off the line with a recent CS/math grad and Ukrainian woman who just got laid off seven months after starting her career in tech. she's on her own here and supporting her family there.

Redoing the #introduction after moving server.

I'm Elias (he/him)

I run a small regenerative farm called Viriditas in #Catalunya (Entenc el català, hablo castillano).

I'm interested in regenerative agriculture, self organisation, federated food security #RegenerativeAg, #Anarchism, #Decentralization.

Especially looking for other regenerative farmers producing in dry climates to discuss best practices.

#Farming #RegenAg #RegenerativeFarmers #RegenerativeAgriculture #NoTill #NoDig

@simonemargio 😛 I’m sorry you had to go through that … I really wish that certain things were better designed instead of just kind of put together as we go along …

@simonemargio Yeah, that’s what it sounds like … What language/platform were you using?

I don’t know if this is the same on other platforms/languages but #Apple #Swift support for image metadata retrieval seems to be rather pitiful 😛 I tried multiple different suggested approaches and lirbaries and all I get are around 4 - 5 meta data items for a particular image.

I use exiftool (https://exiftool.org/) and I get around 20+ metadata items for the same file.

So I resorted to this exiftool wrapper — https://github.com/hlemai/ExifTool It is supposed to be using the exiftool libraries but even that got only 18 items and it left out the one item I was actually interested in 😛

So I finally ended up writing some custom code which would run exiftool locally as a task, get the output of the command and then parse the output to get the metadata. That finally worked for me…

But should I have to go through all these hoops to get image metadata when using Swift? I would have thought there was a simpler/easier solution?

Does anyone know of one?

I’m almost tempted to write my own solution in Swift where I read the image raw data, parse the header and get the metadata. But do I really want to? Probably not …

#Apple #Swift #Images #Metadata #Coding

Gazebo by the Lake
By Lois Mounts

#art #painting #mastoart #flowers #trees #nature

Yesterday's Pratchett novel title prompt was: "Raising Steam".

Here's the thing about the prompt — I generated images on macOS initially and I was happy with the images I was getting since I was getting strange stuff. Nothing really to do with the prompt possibly, but all sorts of weird and wonderful landscapes 🙂

Then I switched to Windows for generation and suddenly all I'd get were trains or some sort of steam engine. Not a lot of variety ... No matter how many models I tried 😛

I've selected a mixed set from both sides for fair representation but I feel as if this needs more exploration ...

#AIArt #StableDiffusion #DeepLearning #MachineLearning #CV #AI #DiscWorld
Prompt: “Raising Steam”. A dark…
Prompt: “Raising Steam”. Some s…
Prompt: “Raising Steam”. A stra…
Prompt: “Raising Steam”. A dark…

And that’s a wrap, folks 🙂 5 papers in the cs.CV category and 4 outside boosted today based on new papers on arXiv.org.

See you tomorrow for the end of the week!

#AI #CV #NewPapers #DeepLearning #MachineLearning

"Forward Pass: On the Security Implications of Email Forwarding Mechanism and Policy" — How email forwarding can create security vulnerabilities and and allow spoofing.

Paper: https://arxiv.org/abs/2302.07287

#AI #NewPaper #Security
Example message with a FROM hea…

"AI Chat Assistants can Improve Conversations about Divisive Topics" — A study looking at how Large Language Models can improve conversations on divisive topics by making the participants feel understood.

Paper: https://arxiv.org/abs/2302.07268

#AI #NewPaper #DeepLearning #MachineLearning #Language #HumanComputerInteraction
Treated Conversation Flow: Resp…

"CiteSee: Augmenting Citations in Scientific Papers with Persistent and Personalized Historical Context" — A tool that uses a reader’s publishing, reading, and saving history to provide personalised context to citations in papers that they’re reading.

Paper: https://arxiv.org/abs/2302.07302

#NewPaper #HumanComputerInteraction
CiteSee augments inline citatio…

"Augmenting Pathologists with NaviPath: Design and Evaluation of a Human-AI Collaborative Navigation System" — Using the navigation strategies of human pathologists to use their domain knowledge to enhance machine learning systems to navigate hight-resolution tumor images to search for patterns of interest.

Paper: https://arxiv.org/abs/2302.07309

#AI #NewPaper #DeepLearning #MachineLearning #HumanComputerInteraction

"Stitchable Neural Networks. (arXiv:2302.06586v2 [cs.LG] UPDATED)" — A way to combine different pretrained models to combine models of varying complexity and performance.

Paper: http://arxiv.org/abs/2302.06586

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Compared with previous scalable…

"Learning When to Say "I Don't Know". (arXiv:2209.04944v2 [cs.CV] UPDATED)" — A method to teach learning systems when they don't know something, or at least to identify areas of uncertainty. Perhaps this should be tried with ChatGPT and Bing to mitigate all the gaslighting? 😛

Paper: http://arxiv.org/abs/2209.04944
Code: https://github.com/osu-cvl/learning-idk

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
t-SNE plots of logits from a we…

"SoK: Anti-Facial Recognition Technology. (arXiv:2112.04558v2 [cs.CR] UPDATED)" — An analysis of the currently available Anti-Facial Recognition (AFR) research and the pros and cons of the different approaches.

Paper: http://arxiv.org/abs/2112.04558

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The workflow of how facial reco…

"Team Triple-Check at Factify 2: Parameter-Efficient Large Foundation Models with Feature Representations for Multi-Modal Fact Verification. (arXiv:2302.07740v1 [cs.CL])" — Verifying facts across different modes (text and images) and types (claim and document).

Paper: http://arxiv.org/abs/2302.07740
Code: https://github.com/wwweiwei/Pre-CoFactv2-AAAI-2023

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
An example of each category in …

"Video Probabilistic Diffusion Models in Projected Latent Space. (arXiv:2302.07685v1 [cs.CV])" — Generating high-resolution and coherent video using diffusion models.

Paper: http://arxiv.org/abs/2302.07685

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
256×256 resolution, 128 frame v…

A total of 54 papers in the cs.CV category on arXiv.org today — 33 new, 21 updated.

Are we slowing down for the weekend? 😛

#AI #CV #NewPapers #DeepLearning #MachineLearning

I do most of my #StableDiffusion image generation using various scripts or GUIs that I created. Each of them store the image generation parameters in the image as meta data.

My latest generation GUI is also an image gallery. It lets me view the images I’ve generated and being able to see the parameters is handy since I can compare various images later and see what model, prompt, and settings were used.

I also do image generation on #Windows (as opposed to #Mac) using #InvokeAI. Now InovkeAI does store the generation settings in the image too but obviously, the keys used to store the data are different to mine.

So my image gallery does not show the settings used when I browse images generated in InokeAI 😛

I got a bunch of coding done and the new GUI working correctly as I liked yesterday. Now I have to figure out how to extract the metadata for InvokeAI-generated images and to display that in my viewer too … It never stops 😀

#Coding #ImageViewer #MachineLearning