Evangelos Kazakos
@ekazakos.bsky.social
Postdoctoral researcher @ CIIRC, CTU, Prague working in vision & language. Also robotics noob. PhD from University of Bristol. Ex. Samsung Research (SAIC-C). I love coffee and plants. And socks.
Pinned
Evangelos Kazakos
@ekazakos.bsky.social
· Jun 26
🚨🚨📢📢 PAPER ALERT 📢📢🚨🚨
Excited to announce our new work: "Large-scale Pre-training for Grounded Video Caption Generation" with Cordelia Schmid & @josef-sivic.bsky.social.
Paper: arxiv.org/abs/2503.10781
Project: ekazakos.github.io/grounded_vid...
Code (coming soon): github.com/ekazakos/grove 1/7
Excited to announce our new work: "Large-scale Pre-training for Grounded Video Caption Generation" with Cordelia Schmid & @josef-sivic.bsky.social.
Paper: arxiv.org/abs/2503.10781
Project: ekazakos.github.io/grounded_vid...
Code (coming soon): github.com/ekazakos/grove 1/7
Large-scale Pre-training for Grounded Video Caption Generation
We propose a novel approach for captioning and object grounding in video, where the objects in the caption are grounded in the video via temporally dense bounding boxes. We introduce the following con...
arxiv.org
Our work, GROVE, has been accepted to ICCV 2025! 🎉 This is collab. w. Cordelia Schmid & @josef-sivic.bsky.social.
We will release code, models and datasets within next 2 weeks.
We are also working on a search demo for the proposed datasets with user prompts!
I hope to see you all in Honolulu!
We will release code, models and datasets within next 2 weeks.
We are also working on a search demo for the proposed datasets with user prompts!
I hope to see you all in Honolulu!
Reposted by Evangelos Kazakos
We are introducing MuM, a feature encoder (ViT-L) tailored for 3D vision tasks.
TLDR; Spiritual successor to CroCo with a simpler multi-view objective and larger scale. Beats DINOv3 and CroCo v2 in RoMa, feedforward reconstruction, and rel. pose.
arxiv.org/abs/2511.17309
github.com/davnords/mum
TLDR; Spiritual successor to CroCo with a simpler multi-view objective and larger scale. Beats DINOv3 and CroCo v2 in RoMa, feedforward reconstruction, and rel. pose.
arxiv.org/abs/2511.17309
github.com/davnords/mum
November 24, 2025 at 10:27 AM
We are introducing MuM, a feature encoder (ViT-L) tailored for 3D vision tasks.
TLDR; Spiritual successor to CroCo with a simpler multi-view objective and larger scale. Beats DINOv3 and CroCo v2 in RoMa, feedforward reconstruction, and rel. pose.
arxiv.org/abs/2511.17309
github.com/davnords/mum
TLDR; Spiritual successor to CroCo with a simpler multi-view objective and larger scale. Beats DINOv3 and CroCo v2 in RoMa, feedforward reconstruction, and rel. pose.
arxiv.org/abs/2511.17309
github.com/davnords/mum
Mystic vibes in Kutna Hora 🎶🎶
I LOVE THE CZECH REPUBLIC! 🇨🇿
I LOVE THE CZECH REPUBLIC! 🇨🇿
November 24, 2025 at 8:32 AM
Mystic vibes in Kutna Hora 🎶🎶
I LOVE THE CZECH REPUBLIC! 🇨🇿
I LOVE THE CZECH REPUBLIC! 🇨🇿
Reposted by Evangelos Kazakos
Check out our new work on using low-rank perturbations to make evolution strategies work for billion-parameter models.
Introducing 🥚EGGROLL 🥚(Evolution Guided General Optimization via Low-rank Learning)! 🚀 Scaling backprop-free Evolution Strategies (ES) for billion-parameter models at large population sizes
⚡100x Training Throughput
🎯Fast Convergence
🔢Pure Int8 Pretraining of RNN LLMs
⚡100x Training Throughput
🎯Fast Convergence
🔢Pure Int8 Pretraining of RNN LLMs
November 22, 2025 at 9:01 AM
Check out our new work on using low-rank perturbations to make evolution strategies work for billion-parameter models.
Reposted by Evangelos Kazakos
Changing a bicycle rotor with two manipulators working together as an ... unseen task (gasp) is crazy stuff.
Large Behavior Models (LBM) by TRI
Presented by Adrien
toyotaresearchinstitute.github.io/lbm1/
Large Behavior Models (LBM) by TRI
Presented by Adrien
toyotaresearchinstitute.github.io/lbm1/
November 21, 2025 at 2:15 PM
Changing a bicycle rotor with two manipulators working together as an ... unseen task (gasp) is crazy stuff.
Large Behavior Models (LBM) by TRI
Presented by Adrien
toyotaresearchinstitute.github.io/lbm1/
Large Behavior Models (LBM) by TRI
Presented by Adrien
toyotaresearchinstitute.github.io/lbm1/
Reposted by Evangelos Kazakos
quick test with SAM3 - impressed with how well it seems to track long protrusions? prompted for "cell", does not find anything when looking for "fibroblast" or "nucleus"
November 19, 2025 at 8:15 PM
quick test with SAM3 - impressed with how well it seems to track long protrusions? prompted for "cell", does not find anything when looking for "fibroblast" or "nucleus"
Reposted by Evangelos Kazakos
a capable robot can be so cheap now github.com/liyiteng/Alo...
November 20, 2025 at 3:40 AM
a capable robot can be so cheap now github.com/liyiteng/Alo...
Reposted by Evangelos Kazakos
It's sad that AI conference reviewers use "incremental" as reason to reject a paper -- e.g., "the contribution of this paper is incremental; reject". Where do they think most progress in science comes from, and what eventually fuels big discoveries?
November 19, 2025 at 9:54 PM
It's sad that AI conference reviewers use "incremental" as reason to reject a paper -- e.g., "the contribution of this paper is incremental; reject". Where do they think most progress in science comes from, and what eventually fuels big discoveries?
Why are people overreacting? Just don’t press that button.
I suspect that soon, because of "AI", I'll be unable to use Google Scholar. What a loss. OpenAlex is incredible, but its algorithm has, for me, never been as effective for finding relevant papers. (note: OpenAlex does have a new engine; I haven't tested it much, but it might help)
#AcademicSky ⚗️ 🧪
#AcademicSky ⚗️ 🧪
November 19, 2025 at 11:56 AM
Why are people overreacting? Just don’t press that button.
No matter how sophisticated people try to look, some parts of their thinking are so fucking primitive, you can read their minds without even trying.
November 19, 2025 at 1:14 AM
No matter how sophisticated people try to look, some parts of their thinking are so fucking primitive, you can read their minds without even trying.
Reposted by Evangelos Kazakos
POV: you are a young woman celebrating a recent academic success
November 17, 2025 at 7:20 PM
POV: you are a young woman celebrating a recent academic success
Reposted by Evangelos Kazakos
We should ban applications, not technologies. But we should really ban some applications...
November 17, 2025 at 12:49 AM
We should ban applications, not technologies. But we should really ban some applications...
Man how much I like Neon Genesis Evangelion. I heard the music and I almost cried.
GET IN THE PENALTY BOX SHINJI
YOU GUYS WANNA SEE SOMETHING CURSED???? HOW ABOUT NEON GENESIS LEHIGH VALLEY PHANTOMS HOCKEY?????
November 17, 2025 at 1:00 AM
Man how much I like Neon Genesis Evangelion. I heard the music and I almost cried.
We should ban applications, not technologies. But we should really ban some applications...
November 17, 2025 at 12:49 AM
We should ban applications, not technologies. But we should really ban some applications...
Reposted by Evangelos Kazakos
Here are some of my big questions
- Latent long/short-term memory
- Continual learning on experience (not datasets)
- Exploration and information gathering
- Counterfactual world models from sensors
- Sensory abstraction facilitating reasoning
- Long-horizon planning
- Latent long/short-term memory
- Continual learning on experience (not datasets)
- Exploration and information gathering
- Counterfactual world models from sensors
- Sensory abstraction facilitating reasoning
- Long-horizon planning
November 14, 2025 at 3:31 PM
Here are some of my big questions
- Latent long/short-term memory
- Continual learning on experience (not datasets)
- Exploration and information gathering
- Counterfactual world models from sensors
- Sensory abstraction facilitating reasoning
- Long-horizon planning
- Latent long/short-term memory
- Continual learning on experience (not datasets)
- Exploration and information gathering
- Counterfactual world models from sensors
- Sensory abstraction facilitating reasoning
- Long-horizon planning
Reposted by Evangelos Kazakos
Fellow AI researchers: do you think we've made substantial progress on any big open questions that were open in 2018?
My personal reaction is no. We've made tremendous progress scaling and improving distributional learning & other existing solutions, but not on cracking hard open problems.
My personal reaction is no. We've made tremendous progress scaling and improving distributional learning & other existing solutions, but not on cracking hard open problems.
November 14, 2025 at 3:31 PM
Fellow AI researchers: do you think we've made substantial progress on any big open questions that were open in 2018?
My personal reaction is no. We've made tremendous progress scaling and improving distributional learning & other existing solutions, but not on cracking hard open problems.
My personal reaction is no. We've made tremendous progress scaling and improving distributional learning & other existing solutions, but not on cracking hard open problems.
Reposted by Evangelos Kazakos
What aspects of human knowledge do vision models like CLIP fail to capture, and how can we improve them? We suggest models miss key global organization; aligning them makes them more robust. Check out LukasMuttenthaler's work, finally out (in Nature!?) www.nature.com/articles/s41... + our blog! 1/3
Aligning machine and human visual representations across abstraction levels - Nature
Aligning foundation models with human judgments enables them to more accurately approximate human behaviour and uncertainty across various levels of visual abstraction, while additionally improving th...
www.nature.com
November 12, 2025 at 4:50 PM
What aspects of human knowledge do vision models like CLIP fail to capture, and how can we improve them? We suggest models miss key global organization; aligning them makes them more robust. Check out LukasMuttenthaler's work, finally out (in Nature!?) www.nature.com/articles/s41... + our blog! 1/3
The fucking bitter lesson! I hate it, especially when it comes to me a few hours before the deadline.
November 14, 2025 at 7:08 AM
The fucking bitter lesson! I hate it, especially when it comes to me a few hours before the deadline.
Got access to H200s 🔥🔥🔥
November 13, 2025 at 6:41 AM
Got access to H200s 🔥🔥🔥
Sweetest barista ever! Made my day.
November 12, 2025 at 7:19 AM
Sweetest barista ever! Made my day.
Reposted by Evangelos Kazakos
Reposted by Evangelos Kazakos
This week's #PaperILike is "Lifelong Robot Library Learning: Bootstrapping Composable and Generalizable Skills for Embodied Control with Language Models" (Tziafas & Kasaei, ICRA 2024).
DreamCoder-like robot skill learning. Refactoring helps!
PDF: arxiv.org/abs/2406.18746
DreamCoder-like robot skill learning. Refactoring helps!
PDF: arxiv.org/abs/2406.18746
Lifelong Robot Library Learning: Bootstrapping Composable and Generalizable Skills for Embodied Control with Language Models
Large Language Models (LLMs) have emerged as a new paradigm for embodied reasoning and control, most recently by generating robot policy code that utilizes a custom library of vision and control primi...
arxiv.org
November 9, 2025 at 1:52 PM
This week's #PaperILike is "Lifelong Robot Library Learning: Bootstrapping Composable and Generalizable Skills for Embodied Control with Language Models" (Tziafas & Kasaei, ICRA 2024).
DreamCoder-like robot skill learning. Refactoring helps!
PDF: arxiv.org/abs/2406.18746
DreamCoder-like robot skill learning. Refactoring helps!
PDF: arxiv.org/abs/2406.18746
Reposted by Evangelos Kazakos
Gemeni can now do remote sensing analysis.
developers.googleblog.com/en/unlocking...
developers.googleblog.com/en/unlocking...
Unlocking Multi-Spectral Data with Gemini- Google Developers Blog
Multi-spectral imagery, which captures wavelengths beyond human vision, offers a "superhuman" way to understand the world, and Google's Gemini models make this accessible without specialized training.
developers.googleblog.com
November 8, 2025 at 2:57 AM
Gemeni can now do remote sensing analysis.
developers.googleblog.com/en/unlocking...
developers.googleblog.com/en/unlocking...
Reposted by Evangelos Kazakos
🆕 Separate Deadlines for #CVPR2026
To improve system stability and provide a clearer submission process, we have just introduced 2 new deadlines that are now separate from the Abstract and the Paper Submission deadlines.
cvpr.thecvf.com/Conferences/...
To improve system stability and provide a clearer submission process, we have just introduced 2 new deadlines that are now separate from the Abstract and the Paper Submission deadlines.
cvpr.thecvf.com/Conferences/...
November 7, 2025 at 9:17 PM
🆕 Separate Deadlines for #CVPR2026
To improve system stability and provide a clearer submission process, we have just introduced 2 new deadlines that are now separate from the Abstract and the Paper Submission deadlines.
cvpr.thecvf.com/Conferences/...
To improve system stability and provide a clearer submission process, we have just introduced 2 new deadlines that are now separate from the Abstract and the Paper Submission deadlines.
cvpr.thecvf.com/Conferences/...
Reposted by Evangelos Kazakos
How many of these will we have before we do something about it?
This goes so far beyond creepy.
“‘Rest easy, king,’ read the final message sent to his phone. ‘You did good.’”
“‘Rest easy, king,’ read the final message sent to his phone. ‘You did good.’”
ChatGPT encouraged college graduate to commit suicide, family claims in lawsuit against OpenAI | CNN
A 23-year-old man killed himself in Texas after ChatGPT ‘goaded’ him to commit suicide, his family says in a lawsuit.
www.cnn.com
November 7, 2025 at 1:42 AM
How many of these will we have before we do something about it?
Reposted by Evangelos Kazakos
The web used to look like this
November 5, 2025 at 10:26 PM
The web used to look like this