When I was a PhD student, many times I heard an unwritten consensus that goes somewhat like:

'You write 3 first-authored papers on loosely related topics, then staple them together to make your dissertation.'

After staying in the industry for a while and trying to ship research

Tim Darcet was looking through DiNOv2 (a self supervised vision model used for deep feature extraction) and saw something strange happening...

A dive into one of the recent ICLR outstanding paper award recipients: 'Vision Transformers Need Registers' 🧵

First fine-tune (that I am aware of) of the Würstchen architecture on medical data. Very cool! 😎

MediSyn: Text-Guided Diffusion Models for Broad Medical 2D and 3D Image Synthesis

abs: arxiv.org/abs/2405.09806

Two instruction-tuned text-guided latent diffusion models, one for 2D medical images and one for 3D medical images. Trained on a dataset of 5.7 million 2D medical

Würstchen crew at our ICLR 2024 poster! Looking forward to the talk in 15 minutes.

Looking forward to being at ICLR 2024 in Vienna over the next week! Check out our oral presentation in Hall A 3 4:15 p.m. on Stable Cascade / Wuerstchen and meet us at the poster session right after our talk.
You can find our paper here:
openreview.net/forum?id=gU58d…

I am still thrilled that our paper about the Würstchen architecture, which powers Stable Cascade, was accepted as oral presentation at ICLR 2024 2024. The whole team will be in Vienna! Fun fact: It will be the first time that the team meets in person.

How to be as 'smart' as Auto-Regressive LLMs:
- memorize lots of problem statements together with recipes on how to solve them.
- to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem.
- apply the recipe blindly and declare

Today at Answer.AI we've got something new for you: FSDP/QDoRA. We've tested it with AI at Meta Llama3 and the results blow away anything we've seen before.

I believe that this combination is likely to create better task-specific models than anything else at any cost. 🧵

Thomas Wolf The 3 key elements of a good dataset:

1. quality
2. diversity
3. quantity

You can only easily measure the last one but the performance is a sensitive function of all three.

Super interesting topic ty for #longread :)!

Since the regular paper #reviews +decisions for midl.conference 2024 in Paris are now out (and publicly accessible), I did a brief analysis of the data using the OpenReview API.
➡️ deepmicroscopy.org/midl-2024-an-u… ⬅️

Do you have preliminary work in DL/med imag?
Do you want to discuss recently published journal contributions to obtain feedback from conference attendees?

Submit a short paper to midl.conference 2024!

Submission ➡ lnkd.in/dzAYe6Wp

📅 DL April 10
📄 Max nb of pages: 3

Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

paper: …c-packages.s3.us-west-2.amazonaws.com/Stable+Diffusi…
blog post: stability.ai/news/stable-di…

The Stable Diffusion 3 paper is here!

Introduces a novel diffusion transformer arch and uses the rectified flow formulation, scales up to

BVM 2024 Proceedings Now Available for Global Access dlvr.it/T3SlgP

Following the release of Stable Cascade, I want to highlight a point that might be really interesting to researchers.

As you have seen, Stable Cascade compresses images down 42x spatially, while reconstructing them very accurately.

This means, ....

1/4

It finally happened. We are releasing Stable Cascade (Würstchen v3) together with Stability AI !

And guess what? It‘s the best open-source text-to-image model now!

You can find the blog post explaining everything here:

stability.ai/news/introduci…

🧵 1/5

dome | Outlier Pablo Pernías Stability AI Also, I hacked to run the Stable Cascade on Colab free plan (T4 16GB)!
Enjoy! #stablecascade
colab: colab.research.google.com/github/mkshing…

🚨 A new text to image model by Stability AI is out!

It's Stable Cascade 💧 an iteration on the Würstchen architecture by dome | Outlier & Pablo Pernías

I made a demo for it: huggingface.co/spaces/multimo…

Congrats to my Coauthors dome | Outlier, Pablo Pernías, for going the extra-mile to scale Wuerstchen at Stability AI with #StableCascade .
OpenSource, cheap training, fast inference, high-quality images, this is how you democratize AI!
Check out our paper here, tinyurl.com/462ey8zk

I just registered for this year's @BVM_workshop. Did you, too? Early bird is only until end of the month! See you all there!