"T2I-Adapter: Learning Adapters to Dig out More Controllable Ability for Text-to-Image Diffusion Models. (arXiv:2302.08453v1 [cs.CV])" — Controlling text-to-image diffusion models in a more granular fashion by using special adapters to provide extra guidance.

Paper: http://arxiv.org/abs/2302.08453
Code: https://github.com/TencentARC/T2I-Adapter

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
We propose T2I-Adapter, a simpl…

"MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation. (arXiv:2302.08113v1 [cs.CV])" — Controlling diffusion-based image generation so that you can specify image components, component placement etc. without any further fine-tuning.

Paper: http://arxiv.org/abs/2302.08113
Code: https://github.com/omerbt/MultiDiffusion

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
MultiDiffusion enables flexible…

"\`A-la-carte Prompt Tuning (APT): Combining Distinct Data Via Composable Prompting. (arXiv:2302.07994v1 [cs.LG])" — Having multiple subsets of data trained on specific prompts and being able to compose the final model based on the prompts you select.

Paper: http://arxiv.org/abs/2302.07994

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
A-la-carte Learning and APT. Gi…

"PRedItOR: Text Guided Image Editing with Diffusion Prior. (arXiv:2302.07979v1 [cs.CV])" — Structure preserving, text guided image editing using diffusion models without needing a base prompt, fine-tuning of models etc.

Paper: http://arxiv.org/abs/2302.07979

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Examples of text guided image e…

@memetzgz If I wanted to read all of them, yes 😛 But I just skim through the summaries to find the ones that look interesting … At the moment, I’ve found 5 interesting papers and have whittled down the 93 to 39 remaining papers …

I saw this little guy yesterday and I can't stop thinking about him. So round. So pink. #borbs #borb

(It's an Australian pink robin and the photos are not mine.)

Credits:
1. © Deepak Karra http://instagram.com/ravi_arora/
2. © Ravi Arora http://instagram.com/ravi_arora/
3. © Ambika Angela Bone http://instagram.com/ambikangela/
4. © Tim J. Hopwood http://flickr.com/photos/timjhopwood/

A total of 93 papers in the cs.CV category on arXiv.org today — 61 new, 32 updated going in to the weekend …

So yesterday’s low paper count was definitely not a slowdown 😛

#AI #CV #NewPapers #DeepLearning #MachineLearning

Updated Akkoma on our server to the latest. New graphs, yay!! Hopefully, nothing broke though since I’m always scared of stuff breaking once I’ve updated since I’m always in a rush …

I ended up testing our new Neural Telephoto feature by shooting with iPhone SE for a while. I loved the shots I got out of it w/ native RAW and the extra reach of the virtual 2×.

Death Valley / Owens Valley, California
iPhone SE, 1× / 2× (Neural Telephoto), @halide RAW

I was quietly minding my own business taking photos of Bullfinches (small/far away) in a tree when this Robin came and perched right in front of my camera and demanded to have his picture taken. How could I refuse?

#AngryBirds 🤣

#Birds #BirdWatching #Twitching #Nature #BirdsOfMastodon #Bird #Photo #Photography #Robin

#FleurisTonFil
La passiflore grande et belle fleur
#photographie #photography

The single-digit temperatures and a bit of sunshine created an ethereal ground fog this morning.

📆 Feb 16, 2023
📷 1/1600 s at f/11, ISO 180, 170mm

#photography #fog #winter #cold #nature #landscape #tree #nikon

The Adam optimizer is at the heart of modern AI. Researchers have been trying to dethrone Adam for years.

How about we ask a machine to do a better job? @googleai uses evolution to discover a simpler & efficient algorithm with remarkable features.

It’s just 8 lines of code: 🧵

@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