Christian Wolf
@chriswolfvision.bsky.social
Principal Scientist at Naver Labs Europe, Lead of Spatial AI team. AI for Robotics, Computer Vision, Machine Learning. Austrian in France. https://chriswolfvision.github.io/www/
Reposted by Christian Wolf
We're pleased to share that the much awaited Anny-One dataset is now available!
Download ➡️ tinyurl.com/anny-one
*⃣ 800K images
🚀Multi-view & multi-person
🏠Indoor 3D scenes
📷Wild cameras
🔥Anny format! github.com/naver/anny
Download ➡️ tinyurl.com/anny-one
*⃣ 800K images
🚀Multi-view & multi-person
🏠Indoor 3D scenes
📷Wild cameras
🔥Anny format! github.com/naver/anny
February 10, 2026 at 11:27 AM
We're pleased to share that the much awaited Anny-One dataset is now available!
Download ➡️ tinyurl.com/anny-one
*⃣ 800K images
🚀Multi-view & multi-person
🏠Indoor 3D scenes
📷Wild cameras
🔥Anny format! github.com/naver/anny
Download ➡️ tinyurl.com/anny-one
*⃣ 800K images
🚀Multi-view & multi-person
🏠Indoor 3D scenes
📷Wild cameras
🔥Anny format! github.com/naver/anny
Is "Bridging the XXX" the new "XXX is all you need" ?
February 10, 2026 at 8:12 AM
Is "Bridging the XXX" the new "XXX is all you need" ?
High-level thoughts on Imitation Learning (Behavior Cloning) vs. Reinforcement Learning for a talk at the winter school this week targeting beginners.
The winter school: project.inria.fr/soraim
The winter school: project.inria.fr/soraim
February 9, 2026 at 8:43 AM
High-level thoughts on Imitation Learning (Behavior Cloning) vs. Reinforcement Learning for a talk at the winter school this week targeting beginners.
The winter school: project.inria.fr/soraim
The winter school: project.inria.fr/soraim
I am in the "Machine Learning Researchers and Scientists" cloud, next to @pierrealquier.bsky.social 😀
This semantic map is nice work and really helpful, thanks!
This semantic map is nice work and really helpful, thanks!
February 9, 2026 at 7:48 AM
I am in the "Machine Learning Researchers and Scientists" cloud, next to @pierrealquier.bsky.social 😀
This semantic map is nice work and really helpful, thanks!
This semantic map is nice work and really helpful, thanks!
I pasted the LaTeX expression of a reward function into Gemini asking it to add curly braces under each term. Of course it did it ... but it also recognized the terms and labelled them correctly, eg. adding "Step Penalty" to \Lambda. Crazy stuff.
Yes, it might be in the training data, but still.
Yes, it might be in the training data, but still.
February 8, 2026 at 10:52 PM
I pasted the LaTeX expression of a reward function into Gemini asking it to add curly braces under each term. Of course it did it ... but it also recognized the terms and labelled them correctly, eg. adding "Step Penalty" to \Lambda. Crazy stuff.
Yes, it might be in the training data, but still.
Yes, it might be in the training data, but still.
C'est la première fois que je vois l'usage du passé simple pour le nom d'un produit ...
February 6, 2026 at 11:58 AM
C'est la première fois que je vois l'usage du passé simple pour le nom d'un produit ...
Reposted by Christian Wolf
We are recruiting four positions connected to Machine Learning, Statistical Learning, and AI for Science in the Applied Mathematics department at École polytechnique. Join our vibrant community at IP Paris and Hi! Paris IA center. List below🧵 tinyurl.com/3jpw9t26
Calliopé
tinyurl.com
February 6, 2026 at 7:56 AM
We are recruiting four positions connected to Machine Learning, Statistical Learning, and AI for Science in the Applied Mathematics department at École polytechnique. Join our vibrant community at IP Paris and Hi! Paris IA center. List below🧵 tinyurl.com/3jpw9t26
Me: 'if you don't propose a topic for our next team building fun event I'll take you to some Commodore 64 exposition or something like that'
People thought that this was some obscure Manga / Anime reference 😩
People thought that this was some obscure Manga / Anime reference 😩
February 5, 2026 at 10:32 PM
Me: 'if you don't propose a topic for our next team building fun event I'll take you to some Commodore 64 exposition or something like that'
People thought that this was some obscure Manga / Anime reference 😩
People thought that this was some obscure Manga / Anime reference 😩
I taught a class yesterday (first lecture in 4y) and students were surprised that shocked that coding w/o an LLM is possible for a human and that they should be capable of solving the assignments w/o LLMs.
If you want to stand out, distinguish yourself from people pasting stuff into an LLM.
If you want to stand out, distinguish yourself from people pasting stuff into an LLM.
February 5, 2026 at 7:02 AM
I taught a class yesterday (first lecture in 4y) and students were surprised that shocked that coding w/o an LLM is possible for a human and that they should be capable of solving the assignments w/o LLMs.
If you want to stand out, distinguish yourself from people pasting stuff into an LLM.
If you want to stand out, distinguish yourself from people pasting stuff into an LLM.
Reposted by Christian Wolf
What if position encodings were designed for vision from scratch? We introduce PaPE—Parabolic Position Encoding. Outperforms RoPE on 7/8 datasets and extrapolates to higher resolutions without fine-tuning or position interpolation. Paper, code, and website in thread 🧵
February 4, 2026 at 8:22 AM
What if position encodings were designed for vision from scratch? We introduce PaPE—Parabolic Position Encoding. Outperforms RoPE on 7/8 datasets and extrapolates to higher resolutions without fine-tuning or position interpolation. Paper, code, and website in thread 🧵
COLLAS added to the AI conf DL countdown:
Abstract: 10.4.2026, AoE
Paper: 15.4.2026, AoE
lifelong-ml.cc/Conferences/...
Abstract: 10.4.2026, AoE
Paper: 15.4.2026, AoE
lifelong-ml.cc/Conferences/...
February 4, 2026 at 1:37 PM
COLLAS added to the AI conf DL countdown:
Abstract: 10.4.2026, AoE
Paper: 15.4.2026, AoE
lifelong-ml.cc/Conferences/...
Abstract: 10.4.2026, AoE
Paper: 15.4.2026, AoE
lifelong-ml.cc/Conferences/...
"Cherry picking" in scientific papers.
(Repost of an older post on some other site in 2021).
(Repost of an older post on some other site in 2021).
February 2, 2026 at 4:17 PM
"Cherry picking" in scientific papers.
(Repost of an older post on some other site in 2021).
(Repost of an older post on some other site in 2021).
A great video about AI-chip supply chains and why TSMC cannot just easily scale up to satisfy the current chip hunger.
TLDR: the long supply chains have complex dynamics and demand spikes are amplified when they travel backwards.
TLDR 2: it's not chips, but power.
www.youtube.com/watch?v=2lLF...
TLDR: the long supply chains have complex dynamics and demand spikes are amplified when they travel backwards.
TLDR 2: it's not chips, but power.
www.youtube.com/watch?v=2lLF...
February 2, 2026 at 12:57 PM
A great video about AI-chip supply chains and why TSMC cannot just easily scale up to satisfy the current chip hunger.
TLDR: the long supply chains have complex dynamics and demand spikes are amplified when they travel backwards.
TLDR 2: it's not chips, but power.
www.youtube.com/watch?v=2lLF...
TLDR: the long supply chains have complex dynamics and demand spikes are amplified when they travel backwards.
TLDR 2: it's not chips, but power.
www.youtube.com/watch?v=2lLF...
It is interesting that US IT tools dominate in most sectors, but not in the tools for the sharing economy, where national tools seem to work well:
France: Leboncoin
Austria: Willhaben
Spain: Wallapop (bought by Naver)
Italy: Subito
Norway: Finn
Finnland: Blocket
etc.
France: Leboncoin
Austria: Willhaben
Spain: Wallapop (bought by Naver)
Italy: Subito
Norway: Finn
Finnland: Blocket
etc.
February 2, 2026 at 8:29 AM
It is interesting that US IT tools dominate in most sectors, but not in the tools for the sharing economy, where national tools seem to work well:
France: Leboncoin
Austria: Willhaben
Spain: Wallapop (bought by Naver)
Italy: Subito
Norway: Finn
Finnland: Blocket
etc.
France: Leboncoin
Austria: Willhaben
Spain: Wallapop (bought by Naver)
Italy: Subito
Norway: Finn
Finnland: Blocket
etc.
this is who runs this account
February 1, 2026 at 10:18 PM
this is who runs this account
Reposted by Christian Wolf
I know another, well tested, way of facilitating faithful scientific publications: WRITE IT YOURSELF.
February 1, 2026 at 8:50 AM
I know another, well tested, way of facilitating faithful scientific publications: WRITE IT YOURSELF.
I have no idea whether this Moltbot thing will lead to any useful outcomes but it is certainly a good idea to test whether it will... And this will be a useful outcome in itself. So, yes actually.
This is one reason why moltbook is so significant.
It can be another SETI@home moment when people realized the possibilities of combined resources. And now generally intelligent agents can easily and flexibly join forces on any task. And there can be many wisdom of the crowd type multipliers.
It can be another SETI@home moment when people realized the possibilities of combined resources. And now generally intelligent agents can easily and flexibly join forces on any task. And there can be many wisdom of the crowd type multipliers.
February 1, 2026 at 1:12 PM
I have no idea whether this Moltbot thing will lead to any useful outcomes but it is certainly a good idea to test whether it will... And this will be a useful outcome in itself. So, yes actually.
Update to my earlier post: functional meanings can be exploited to align different signals to have the same appearance for the same functional meaning. In Austria, road side reflectors are white on the left side ("pass it on the right side") and the red on the right side ("pass it on the left").
January 30, 2026 at 12:44 PM
Update to my earlier post: functional meanings can be exploited to align different signals to have the same appearance for the same functional meaning. In Austria, road side reflectors are white on the left side ("pass it on the right side") and the red on the right side ("pass it on the left").
Different levels of how an agent understands a scene.
January 29, 2026 at 10:17 PM
Different levels of how an agent understands a scene.
I got long email with research questions, full with long equations written in LaTeX. So instead of compiling this with LaTeX, I dumped it into Gemini, asking it to render the equations for better readability.
Gemini did it ... and started to answer the questions, which it was not asked to do 😂.
Gemini did it ... and started to answer the questions, which it was not asked to do 😂.
January 29, 2026 at 7:18 AM
I got long email with research questions, full with long equations written in LaTeX. So instead of compiling this with LaTeX, I dumped it into Gemini, asking it to render the equations for better readability.
Gemini did it ... and started to answer the questions, which it was not asked to do 😂.
Gemini did it ... and started to answer the questions, which it was not asked to do 😂.
Donald Trump gave the biggest gift to Europe: a violent drive and motivation for digital sovereignty. He couldn't have harmed the US digital hegemony more even if he tried.
January 28, 2026 at 10:31 AM
Donald Trump gave the biggest gift to Europe: a violent drive and motivation for digital sovereignty. He couldn't have harmed the US digital hegemony more even if he tried.
I just read a CVPR 2025 paper which explains that one of the contributions does not increase val/test performance and actually even decreases it, but the training curves show that training error is lower, so everything is fine and this explains the effectiveness of this loss.
(reposted after error)
(reposted after error)
January 27, 2026 at 12:10 PM
I just read a CVPR 2025 paper which explains that one of the contributions does not increase val/test performance and actually even decreases it, but the training curves show that training error is lower, so everything is fine and this explains the effectiveness of this loss.
(reposted after error)
(reposted after error)
If the EU is really serious about IT alternatives, why don't we build a European social network based on atproto? Interfaceable with Bluesky, the users of the new network would see Bluesky posts (and vice versa), data of the new SN would be held in EU servers etc. Did I misunderstand how this works?
January 26, 2026 at 7:34 PM
If the EU is really serious about IT alternatives, why don't we build a European social network based on atproto? Interfaceable with Bluesky, the users of the new network would see Bluesky posts (and vice versa), data of the new SN would be held in EU servers etc. Did I misunderstand how this works?
Reposted by Christian Wolf
This reddit post made me laugh
January 26, 2026 at 10:01 AM
This reddit post made me laugh