Héctor Laria

Postdoc in Computer Vision. GenAI.

[Héctor Laria @hector_laria]

@joanrod_ai @QuiverAI @a16z 🔥🔥 congrats, lots of hard work paying off and more to come!

[Joan Rodriguez @joanrod_ai]

Introducing @QuiverAI, a new AI lab and product company focused on frontier vector design.

We’ve raised an $8.3M seed round led by @a16z, with support from amazing angels and investors.

Our first model, Arrow-1.0, generates SVGs from images and text. It’s available now in public beta at https://t.co/zjAnKlI8pp

[Minhyuk Sung @MinhyukSung]

🎓 Just wrapped up this semester’s “Diffusion and Flow Models” course at KAIST, covering DDPM/DDIM to SDE/ODE and the connection between diffusion and flow models. Next week we host guest talks by Subham Sahoo and Philipp Henzler.
🌐 https://t.co/fYoEth6iM8

[Joan Rodriguez @joanrod_ai]

I’m excited to announce that 💫StarVector has been accepted at CVPR 2025! Over a year in the making, StarVector opens a new paradigm for Scalable Vector Graphics (SVG) generation by harnessing multimodal LLMs to generate SVG code that aesthetically mirrors input images and text. With this milestone, we’re also releasing StarVector on Hugging Face! 🥳🚀

[Alexandra Gomez-Villa 🏳️‍🌈 @alviur]

🔍 Fascinating discovery: Diffusion models perceive light/color as humans do!

In our recently accepted #CVPR2025 @CVPR paper with @Alvaro_kai93 , We've found that during inversion, diffusion models naturally replicate human responses to color visual illusions.

🧵

[Joan Rodriguez @joanrod_ai]

🎉 Excited to introduce BigDocs!
An open, transparent multimodal dataset designed for:

📄 Documents
🌐 Web content
🖥️ GUI understanding
👨‍💻 Code generation from images

We’re also launching BigDocs-Bench, featuring 10 tasks to test models on:

➡️ Document, Web, GUI Visual reasoning
➡️ Converting images into JSON, Markdown, LaTeX, SVG, and more!

📜 Paper: https://t.co/NsXtsIennh https://t.co/EXhEFnQ622
🌍 Website https://t.co/kA9c6JH7L8

[Héctor Laria @hector_laria]

@joanrod_ai @serra9lozano @iclr_conf ours* , we always answer! 😆

[Héctor Laria @hector_laria]

@joanrod_ai @serra9lozano @iclr_conf two of us never answered. one of them asking about the figures, answers in the captions.

[Sihyun Yu @sihyun_yu]

Introducing REPA! We show that learning high-quality representations in diffusion transformers is crucial for boosting generation performance. With REPA, we speed up SiT training by 17.5x (without CFG) and achieve state-of-the-art FID = 1.42 using CFG with the guidance interval. 🧵[1/7]

https://t.co/MoN9vjZ93U

[Stanford AI Lab @StanfordAILab]

arXiv -> alphaXiv

Students at Stanford have built alphaXiv, an open discussion forum for arXiv papers. @askalphaxiv

You can post questions and comments directly on top of any arXiv paper by changing arXiv to alphaXiv in any URL!

[Christian Wolf (🦋🦋🦋) @chriswolfvision]

This is on of the best papers I have recently read, congrats for this excellent work.

"NNs do not have an inherent “simplicity bias”. This property depends on components such as ReLUs, residual connections, and layer normalizations".

This also applies to transformers.

[Jack Morris @jxmnop]

one of the most important things I know about deep learning I learned from this paper: "Pretraining Without Attention"

this what I found so surprising:
these people developed an architecture very different from Transformers called BiGS, spent months and months optimizing it and training different configurations, only to discover that at the same parameter count, a wildly different architecture produces identical performance to transformers

this may imply that as long as there are enough parameters, and things are reasonably well-conditioned (i.e. a decent number of nonlinearities and and connections between the pieces) then it really doesn't matter how you arrange them, i.e. any sufficiently good architecture works just fine

i feel there's something really deep here, and we may be already very close to the upper bound of how well we can approximate a given function given a certain amount of compute. so we should spend more time thinking about other questions, such as what that function should actually look like (what data? which objective function?) and how to make it more efficient

[Héctor Laria @hector_laria]

Getting fond of @dlbcnai.
It's like that cool uncle that comes home for Christmas and tells you amazing stories 🎄

Thanks to the organizers and chairs.

[Héctor Laria @hector_laria]

Thanks everyone, I feel appreciated for getting "best presentation" again!

I hope you found my failures interesting, I certainly learned from them.

[CVC_UAB @CVC_UAB]

And the winner is...

🏆@hector_laria for the best presentation (2nd year in a row)
🏆 Pau Torras for the best quiz score (2nd year in a row)
🏆 Khanh Nguyen for the best poster

Congratulations!!👏🎉

#CVCpeople #phdlife

[Héctor Laria @hector_laria]

@PDillis @quasimondo Related papers are Latent Diffusion Models and Arno's group paper on heat dissipation models, where blur is used instead of noise, and the initial color mass is kept for the final image and just redistributed 🧐 https://t.co/Wxc5mxogmb

[Héctor Laria @hector_laria]

@PDillis @quasimondo Generally, domain specific knowledge is sideloaded during diffusion, i.e., controllable score-based models. But hey what do I know, you could find something interesting!

[Héctor Laria @hector_laria]

@ArashVahdat @RuiqiGao @karsten_kreis Thanks for the extra mile and good luck finishing the survey

[Arash Vahdat @ArashVahdat]

📢📢 The recording of our tutorial on denoising diffusion models is now available on YouTube:
https://t.co/JB7SLuvxuW

Slides and additional info:
https://t.co/5H4zFeovF2

This tutorial was originally presented by @RuiqiGao, @karsten_kreis, and myself at #CVPR2022.