Lightnews — Scholar-powered news

Hilde Kuehne @hildekuehne.bsky.social · 2d

Nice… need to check if the representation learning community found a similar signal…

3

Reposted by Hilde Kuehne

Evangelos Kazakos @ekazakos.bsky.social · 2d

To keep the tradition, the lineup is 🔥🔥🔥

Giorgos Tolias @gtolias.bsky.social · 2d

The Visual Recognition Group at CTU in Prague organizes the 50th Pattern Recognition and Computer Vision Colloquium with
Torsten Sattler, Paul-Edouard Sarlin, Vicky Kalogeiton, Spyros Gidaris, Anna Kukleva, and Lukas Neumann.
On Thursday Oct 9, 11:00-17:00.

cmp.felk.cvut.cz/colloquium/

3 9

Hilde Kuehne @hildekuehne.bsky.social · 2d

is there a paper I missed?

1

Hilde Kuehne @hildekuehne.bsky.social · 2d

🥳

1

Reposted by Hilde Kuehne

Giorgos Tolias @gtolias.bsky.social · 2d

The Visual Recognition Group at CTU in Prague organizes the 50th Pattern Recognition and Computer Vision Colloquium with
Torsten Sattler, Paul-Edouard Sarlin, Vicky Kalogeiton, Spyros Gidaris, Anna Kukleva, and Lukas Neumann.
On Thursday Oct 9, 11:00-17:00.

cmp.felk.cvut.cz/colloquium/

2 6 21

Hilde Kuehne @hildekuehne.bsky.social · 2d

It looks so good… 😋… I‘m trying more veggies, no sweets but I‘m failing all the time

1

Reposted by Hilde Kuehne

AI Conference DL Countdown @dlcountdown.bsky.social · 3d

CVPR'26 (abs): 33 days.
CVPR'26 (paper): 40 days.
ECCV'26: 152 days.

1 5

Reposted by Hilde Kuehne

Serge Belongie @serge.belongie.com · 5d

@belongielab.org welcomes applications for a PhD position in CV/ML (fine-grained analysis of multimodal data, 2D/3D generative models, misinformation detection, self-supervised learning) / Apply though the ELLIS portal / Deadline 31-Oct-2025

ELLIS @ellis.eu · 7d

🎓 The application portal for the #ELLISPhD Program is open today! Info & link: https://bit.ly/45DSe75

ELLIS PhD Program: Call for Applications 2025

The ELLIS mission is to create a diverse European network that promotes research excellence and advances breakthroughs in AI, as well as a pan-European PhD program to educate the next generation of AI...

ellis.eu

9 23

Reposted by Hilde Kuehne

#ICCV2025 @iccv.bsky.social · 11d

If you have an accepted paper at the main #ICCV2025 conference, please upload a thumbnail, a full poster image, and a 5 minute YouTube video presentation for your paper.

Please have the uploads done by October 8, 2025 aoe.

Instructions: iccv.thecvf.com/Conferences/...

1 4 6

Reposted by Hilde Kuehne

#ICCV2025 @iccv.bsky.social · 9d

Hi all! Reminder that the #ICCV2025 Main Conference Poster Art And Video Upload Deadline is coming up on Oct 09 '25 04:00 AM UTC

Clock is ticking for the conference itself as well and we are very excited to see you all there 🥳

3 4

Reposted by Hilde Kuehne

AI Conference DL Countdown @dlcountdown.bsky.social · 5d

CVPR'26 (abs): 35 days.
CVPR'26 (paper): 42 days.
ECCV'26: 154 days.

1 3

Hilde Kuehne @hildekuehne.bsky.social · 5d

But this is not really something which is quantitatively shown in the paper …

1 1

Hilde Kuehne @hildekuehne.bsky.social · 5d

No comment on the second one ... but there should be a rule that all flavours of "emerging" or "emergence" are only allowed in context of models where the complete training data (I mean for the full pipeline) is public.

Everything else should be considered false marketing until proven otherwise.

a cartoon of a girl holding a bag and pointing at a woman

ALT: a cartoon of a girl holding a bag and pointing at a woman

media.tenor.com

1 4

Hilde Kuehne @hildekuehne.bsky.social · 5d

But what we know is that data is full of those basic samples ... just crawl enough conference talks and I show you emergence. And going from low to high res? ... never seen a video like that! And we didn't even touch style transfer.
So nope! But please, help me find the substance here. 4/4

1 2

Hilde Kuehne @hildekuehne.bsky.social · 5d

Second, we know there is tons of data and task leakage. So, without knowing the training data (come on everybody, sharing is caring!), we can not say anything about new "emerge whatever" capabilities (which have been rarely shown for real up to this point).
3/4

1 7

Hilde Kuehne @hildekuehne.bsky.social · 5d

Can you find any serious evaluation in this paper on any standard benchmark and metric for any task they claim? The main evaluation is "We run it ten times and test if it works in one of those cases". Would you accept this evaluation from any computer vision MSc student? Or run a car on that? 2/4

1 3

Hilde Kuehne @hildekuehne.bsky.social · 5d

omg ... where should I start, where should I end ... that rant would go on forever. Probably, to get a glimpse, read it as a research paper ... I mean, at least it was published on arxiv, so we should treat it with minimum respect. Let's go ... 1/4

And please don't take the rant too serious...

1 5

Hilde Kuehne @hildekuehne.bsky.social · 5d

Overhyped….

1 3

Reposted by Hilde Kuehne

Paul Gavrikov @paulgavrikov.bsky.social · 7d

🚨 New paper out!
"VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes"
👉 arxiv.org/abs/2509.25339
We test 37 VLMs on 2,700+ VQA questions about dense scenes.
Findings: even top models fumble badly—<20% on the hardest split and key failure modes in counting, OCR & consistency.

1 2 6

Reposted by Hilde Kuehne

AI Conference DL Countdown @dlcountdown.bsky.social · 7d

CVPR'26 (abs): 37 days.
CVPR'26 (paper): 44 days.
ECCV'26: 156 days.

1 3

Reposted by Hilde Kuehne

ML for Science @ml4science.bsky.social · 7d

We're back at Neckawa in Tübingen today for Day 2 of our Cluster Conference #MLinScience2025! Our first speaker is @hildekuehne.bsky.social with her talk "Advances in Self-Supervised Multimodal Learning". Today's program: uni-tuebingen.de/en/165313#c2...

Scientist Hilde Kühne stands on a stage and gives a talk. On the wall behind her are slides with images illustrating her talk.

1 10

Hilde Kuehne @hildekuehne.bsky.social · 13d

Strands #571
“Simply divine”
🔵🔵🔵🟡
🔵🔵

Hilde Kuehne @hildekuehne.bsky.social · 13d

Connections
Puzzle #837
🟨🟨🟨🟨
🟩🟩🟩🟩
🟦🟦🟦🟦
🟪🟪🟪🟪

Reposted by Hilde Kuehne

Kosta Derpanis @csprofkgd.bsky.social · 14d

At #ICCV2025 we’re hosting the 1st workshop on Geometry-Free Novel View Synthesis & Controllable Video Models! (Or simply: “3D Vision in the era of Video Models”)

Our speaker lineup is 🔥🔥🔥 Better grab a seat early 😉

geofreenvs.github.io

4 25

Reposted by Hilde Kuehne

David Picard @davidpicard.bsky.social · 22d

I'm currently attending the video Symposium sites.google.com/view/video-a...
Starting with already impressively good talks 🤩

2 3 25