iiinova — where your digital self keeps evolving

Reposted by Chris Offner

Yining Karl Li @yiningkarlli.bsky.social · 29/03/2026

Huge props to Lord and Miller for stepping up and doing what directors like Nolan are too cowardly to do: be up front and give loud and generous credit to their VFX team. Great directors don’t need to lie about how their movies are made; the work speaks for itself.

1223

Chris Offner @chrisoffner3d.bsky.social · 29/03/2026

Any idea why Scholar Inbox cannot find the paper arxiv.org/abs/2512.11508? @si-cv-graphics.bsky.social @andreasgeiger.bsky.social

120

Reposted by Chris Offner

Ryan Moulton @moultano.bsky.social · 17/03/2026

If you want me to consider reading something, you have to convince me that you care way more than I do.

1746

Reposted by Chris Offner

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 14/02/2026

All the "you need to learn AI skills or you'll get left behind" things are patently nonsense. It's easy to use and only becomes easier to use over time. If there's skill it's in knowing what it does well and what is does poorly

429817

Reposted by Chris Offner

Tim Crane @timcrane.bsky.social · 11/12/2025

This is a really good point. While there are men and women on the 'sceptic' side of all these debates, I don't know of any women on the AI 'booster' side. It's really a guy thing

5272

Reposted by Chris Offner

Eric Brachmann @ericbrachmann.bsky.social · 17/10/2025

Looking forward to a busy #ICCV2025. I will give three (very different) talks at workshops and tutorials, see info below. We also present two papers, ACE-G and SCR Priors. And it's the 10th (!) anniversary of the R6D workshop, which we co-organize.

1124

Reposted by Chris Offner

Andreas Geiger @andreasgeiger.bsky.social · 01/10/2025

#TTT3R: 3D Reconstruction as Test-Time Training TTT3R offers a simple state update rule to enhance length generalization for #CUT3R — No fine-tuning required! 🔗Page: rover-xingyu.github.io/TTT3R We rebuilt @taylorswift13’s "22" live at the 2013 Billboard Music Awards - in 3D!

0384

Reposted by Chris Offner

European Commission @ec.europa.eu · 06/09/2025

🚀 Europe’s first exascale supercomputer is here! JUPITER, launched in Germany, is the EU’s most powerful system and fourth fastest worldwide. 100% powered by renewables, it has also ranked first in energy efficiency. It will boost AI, science, and climate research. Read more - europa.eu/!vcWBqW

1122652

Reposted by Chris Offner

Ryan Moulton @moultano.bsky.social · 05/09/2025

There is a lot to hate about the politics of the silicon valley right, but they do actually want to build stuff, and I would prefer if the left didn't cede "we should be able to build stuff" to the right.

27426416

Chris Offner @chrisoffner3d.bsky.social · 05/09/2025

People often use "smart" when they mean "wise" and I don't think it's too controversial to doubt the wisdom of some tech elites. Other than that I certainly agree with you.

140

Reposted by Chris Offner

Keenan Crane @keenancrane.bsky.social · 29/08/2025

I can't* fathom why the top picture, and not the bottom picture, is the standard diagram for an autoencoder. The whole idea of an autoencoder is that you complete a round trip and seek cycle consistency—why lay out the network linearly?

1115925

Chris Offner @chrisoffner3d.bsky.social · 23/08/2025

Great video on the convergent evolution from hierarchical military command structures to cybernetics to centralized AI coordination across political ideologies: www.youtube.com/watch?v=mayo...

120

Chris Offner @chrisoffner3d.bsky.social · 22/08/2025

I'd also welcome a Bayesian framing. I know Andrew Davison's group has done work on Gaussian belief propagation for SLAM factor graphs (gaussianbp.github.io) but other than that and arxiv.org/abs/1703.04977, I'm not aware of of much Bayesian (deep) learning in (3D) vision right now.

gaussianbp.github.io

Gaussian Belief Propagation

040

Reposted by Chris Offner

Johan Edstedt @parskatt.bsky.social · 22/08/2025

In general I think 3D vision would do well to take some inspiration from Bayesians. I guess these days they lost their glamour, but imo it's a very nice way of thinking that feels somewhat lost currently.

221

Chris Offner @chrisoffner3d.bsky.social · 22/08/2025

"It is beautiful. It is elegant. Does it work well in practice? Not really. This is often the caveat we face in research: the things that are beautiful don't work and the things that work are not beautiful." – Daniel Cremers

2365

Chris Offner @chrisoffner3d.bsky.social · 22/08/2025

You follow him. Andrew Davison from Imperial College London.

100

Chris Offner @chrisoffner3d.bsky.social · 21/08/2025

"As roboticists and computer vision people [outside of big tech], do we have to just wait for the next foundation model?" I share the frustration. It's disempowering when most major progress recently is downstream of "foundation models" that you don't have the compute or data to train yourself.

5242

Reposted by Chris Offner

Bibliome @handle.invalid · 20/08/2025

We're live on bluesky! bibliome.club is the platform for creating, collaborating on and sharing reading lists with your Bluesky network - open source and decentralised via ATProto.

bibliome.club

Bibliome - Building the very best reading lists, together

Create collaborative bookshelves, discover new books, and build reading communities with friends. Join the decentralized reading revolution powered by Bluesky.

724270

Chris Offner @chrisoffner3d.bsky.social · 19/08/2025

Sort of, but DINOv3 also seems to (inadvertently?) point towards the limits of pure scaling. x.com/chrisoffner3...

230

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

If you maximize cosine similarity, aren't you left with only a single dimension (i.e. scaling the vector norm) as CosSim-invariant "wiggle room" to encode geometric information that isn't also captured by the language?

000

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

Yes but that's an additional training objective beyond merely minimizing cosine similarity. You'd need to introduce something that ensures that pixel features don't just collapse to language semantics, via some auxiliary task, no?

100

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

It just seems to me that mapping pixels and language to highly similar internal representations means that you'll drop a lot of information that is not (or cannot) be accurately described by language.

310

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

If we try to perfectly reconstruct, e.g., a complex 3D mesh from a natural language description, we'll find that the two modalities operate on very different levels of precision and abstraction.

000

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

My concern is that language as a modality inherently biases the data towards coarser labels/concepts. You won't perfectly describe per-pixel normals and depth in natural language. Geometry is continuous and "raw", language is discrete and abstract.

220

Chris Offner @chrisoffner3d.bsky.social · 15/08/2025

Oh, interesting. I'll check that out!

000

Chris Offner @chrisoffner3d.bsky.social · 14/08/2025

Yay, DINOv3 is out! SigLIP (VLMs) and DINO are two competing paradigms for image encoders. My intuition is that joint vision-language modeling works great for semantic problems but may be too coarse for geometry problems like SfM or SLAM. Most animals navigate 3D space perfectly without language.

1315

Chris Offner @chrisoffner3d.bsky.social · 12/08/2025

What are the best resources to learn about VLMs? Papers, tutorials, courses, blog posts, whatever is good. I can read the Kimi-VL or GLM tech reports and follow the breadcrumbs but I'd appreciate any and all recommendations towards a useful VLM curriculum! 🙏

182

Reposted by Chris Offner

Siobhán @shibbi.me · 28/06/2025

Ensuring the robots can’t take our jobs by teaching the robots functional programming

0113

Reposted by Chris Offner

Eric Brachmann @ericbrachmann.bsky.social · 23/06/2025

I was not aware that the #ECCV2024 oral recordings are publicly available... So here is the #ACEZero talk: eccv.ecva.net/virtual/2024...

3282

Chris Offner @chrisoffner3d.bsky.social · 23/06/2025

I don't think having basic type checks is equivalent to doing proofs in Lean. I just don't want to chase through fifty layers of inheritance to figure out what type of input the blessed authors of some godforsaken research codebase expect to be given to their function.

100

Chris Offner @chrisoffner3d.bsky.social · 23/06/2025

Which is why I hope that Mojo will one day (when it's mature enough to be open sourced) save us all and give us a properly typed AI language.

100

Chris Offner @chrisoffner3d.bsky.social · 23/06/2025

I must note that, while the typing is good, the formatting in the above jaxtyping example is still "not hehe." x.com/chrisoffner3...

020

Chris Offner @chrisoffner3d.bsky.social · 23/06/2025

Agreed, but the remedy for bad type hints is not no type hints but good type hints. github.com/patrick-kidg...

340

Reposted by Chris Offner

Christian Wolf @chriswolfvision.bsky.social · 06/06/2025

DUNE is a great universal image encoder which works for a large class of tasks and beats its teacher MASt3R on map free localization. #cvpr2025 We use it a lot already, I recommend it. (keynote @ Paris by @dlarlus.bsky.social )

2435

Reposted by Chris Offner

Bart Wronski 🇺🇦🇵🇸 @bartwr.bsky.social · 01/06/2025

I agree 100%. It's one thing to criticize corporate practices, the social impact, ethics, or future risks. But I watch in total awe how it writes in a few seconds a well documented program in a language/API I don't know, while they complain "but it might have a bug and requires a pass or two" O_o

gilbetron 60 days ago | next [=]
I get so confused on this. I play around, test, and mess with LLMs all the time and they are miraculous. Just amazing, doing things we dreamed about for decades. I mean, I can ask for obscure things with subtle nuance where I misspell words and mess up my question and it figures it out. It talks to me like a person. It generates really cool images. It helps me write code. And just tons of other stuff that astounds me.
And people just sit around, unimpressed, and complain that ... what ... it isn't a perfect superintelligence that understands everything perfectly? This is the most amazing technology I've experienced as a 50+ year old nerd that has been sitting deep in tech for basically my whole life. This is the stuff of science fiction, and while there totally are limitations, the speed at which it is progressing is insane. And people are like, "Wah, it can't write code like a Senior engineer with 20 years of experience!"
Crazy.

9344

Chris Offner @chrisoffner3d.bsky.social · 02/06/2025

Similar to “What’s up? - Not much.” in (British) English I guess.

020

Chris Offner @chrisoffner3d.bsky.social · 31/05/2025

My X feed's reaction to Dario Amodei's recent interviews.

020

Chris Offner @chrisoffner3d.bsky.social · 30/05/2025

"awesome" good "awful" bad 🤷‍♂️

150

Reposted by Chris Offner

Wesley Finck @wesleyfinck.org · 30/05/2025

Collective intelligence is only as strong as our collective attention.

093

Chris Offner @chrisoffner3d.bsky.social · 22/05/2025

Yeah but all my notes and highlights and paper library is in Zotero.

000

Chris Offner @chrisoffner3d.bsky.social · 21/05/2025

Yeah, unfortunately. I'm using the (Chromium-based) Arc browser since two years or so, but since Arc has been abandoned by its developers I'll probably switch to Zen browser (Firefox fork with Arc's superior UI) soon, so will have to give up on that extension. zen-browser.app

zen-browser.app

Zen Browser

Beautifully designed, privacy-focused, and packed with features.

010

Chris Offner @chrisoffner3d.bsky.social · 21/05/2025

I also use Zotero for most of my reading. Other than that, the Google Scholar PDF Reader extension is the best thing w.r.t how citations are handled: chromewebstore.google.com/detail/googl...

chromewebstore.google.com

Google Scholar PDF Reader - Chrome Web Store

Supercharge your paper reading: follow references, skim outline, jump to figures, cite and save.

211

Chris Offner @chrisoffner3d.bsky.social · 20/05/2025

Depending on who you talk to, a “deep dive” can mean spending seven years leading a field of research, or it can mean having six words on a slide instead of two.

292