Lazaros Nalpantidis

Where to Attend: A Principled Vision-Centric Position Encoding with Parabolas We propose Parabolic Position Encoding (PaPE), a parabola-based position encoding for vision modalities in attention-based architectures. Given a set of vision tokens-such as images, point clouds, vid...

Where to Attend: A Principled Vision-Centric Position Encoding with Parabolas

Paper: arxiv.org/abs/2602.01418
Website: chrisohrstrom.github.io/parabolic-po...
Code: github.com/DTU-PAS/para...

@rgring.bsky.social @lanalpa.bsky.social

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 🧵

Kudos to the authors: Junru Ren, Abhijoy Mandal, Rama El-khawaldeh, Shi Xuan Leong, @profhein.bsky.social, @aspuru.bsky.social, @lanalpa.bsky.social and Kourosh Darvish.
[6/6]

This approach enables more reliable real-time monitoring for automated workflows such as liquid–liquid extraction, distillation, and crystallization—bringing us closer to truly adaptive, autonomous chemistry labs.
[5/6]

Now in Digital Discovery: Context-aware computer vision for chemical reaction state detection.

🔗 pubs.rsc.org/en/content/a...
[1/6]

Exciting new research from our group! 🦾🤓🎓

Great talk! Very thought provoking!

Postdoc in Robotics and Computer Vision for Life Science Laboratory Automation - DTU Electro As part of a joint research collaboration between DTU and Novo Nordisk, we are looking for a postdoc to join our multidisciplinary research program focusing on the interplay between AI-Protein design,...

Join us and revolutionize Life Science Lab Automation! 🎓🤖💉

I am hiring a Postdoc in Robotics and Computer Vision for Life Science Laboratory Automation, in Copenhagen, Denmark.

Is that you? 🙋‍♀️
efzu.fa.em2.oraclecloud.com/hcmUI/Candid...

This is part of a strategic partnership between DTU - Technical University of Denmark and Novo Nordisk to tackle one of life science industry's biggest challenges: Closed-Loop Design and Optimization of Biologics.

The goal of this position is to work as part of a larger cross-disciplinary team and contribute by enhancing the visual perception capabilities of robots, enabling them to interact autonomously with unknown and dynamic objects.

Postdoc in Autonomous Robots and Perception for Protein Design Laboratories - DTU Electro As part of a joint research collaboration between DTU and Novo Nordisk, we are looking for a postdoc to join our multidisciplinary research program focusing on the interplay between AI-Protein design,...

I am hiring a Postdoctoral Researcher to work on Computer Vision for Autonomous Robots in Life Science Automation.

#ComputerVision #Robotics #LabAutomation

Apply here: efzu.fa.em2.oraclecloud.com/hcmUI/Candid...

** update: paper accepted in ICRA 2025 **

If you are attending #ICRA2025 in Atlanta, we will be happy to meet you and discuss about our SteeredMarigold method!!

For us, it is an important step in our RoBétArmé Horizon Europe project, funded by @ec.europa.eu, improving the semantic understanding capabilities of our robotic system.

Special thanks to our project partner Christiansen og Essenbæk A/S for organizing construction site access!

We hope our data is an accelerator for everyone training and deploying deep learning models in construction – We think everyone faced the problem of limited data availability, which we hope to alleviate.

This journal article introduces a new publicly available dataset captured in construction environments. We have around 15k images, a mix of self-captured and publicly available images with segmentation labels for defects in reinforced concrete.

GitHub - DTU-PAS/ConRebSeg: ConRebSeg: A Segmentation Dataset for Reinforced Concrete Construction ConRebSeg: A Segmentation Dataset for Reinforced Concrete Construction - DTU-PAS/ConRebSeg

📣NEW DATASET ALERT! 🧱🚧

We are happy to present our latest research work in @elsevierconnect.bsky.social "Automation in Construction" journal: doi.org/10.1016/j.au...
a work driven by my PhD student, Patrick Schmidt!
Find the corresponding repo here: github.com/DTU-PAS/ConR...

Depth completion for real-world depth sensors? 🤖
Check out our latest work: steeredmarigold.github.io

A Survey on Dynamic Neural Networks: from Computer Vision to Multi-modal Sensor Fusion Model compression is essential in the deployment of large Computer Vision models on embedded devices. However, static optimization techniques (e.g. pruning, quantization, etc.) neglect the fact that d...

Our paper “A Survey on Dynamic Neural Networks: from Computer Vision to Multi-modal Sensor Fusion" is out as preprint!

By myself, @sscardapane.bsky.social, @rgring.bsky.social and @lanalpa.bsky.social

📄 arxiv.org/abs/2501.07451

@tpjenkins.bsky.social

Can Dynamic Neural Networks boost Computer Vision and Sensor Fusion?
We are very happy to share this awesome collection of papers on the topic!

Dynamic NNs can significantly improve the perception capabilities of resource-constraint platforms, such as robots! I am looking forward to seeing what is possible!!

[Last Chance]
Two more days before the DL for the Lecturer in Computer Vision Post @bristoluni.bsky.social [6 Jan DL]
This is a great opportunity to establish your own research group in a supportive environment. Pls circulate and apply.
www.jobs.ac.uk/job/DKR225/l...

Of course, we traveled a lot, and of course, we produced great research!
I am wondering what 2025 can bring to us! 🤓

📚 We had great gatherings and academic discussions throughout the year, the best of which was hosted by our friends and colleagues at @aicentre.dk Pioneer Centre for AI !

🏆 Ronja Güldenring @rgring.bsky.social was awarded the prestigious "Young Researcher Award" from DTU - Technical University of Denmark for her PhD thesis!

The academic journey is wonderful when you share it with a great team!

🎓 I am very proud of my 2 PhD students who graduated in 2024, Dimitrios Arapis (now with Novo Nordisk) and Ronja Güldenring (continuing with us in DTU Electro).

Computer Vision: Fact & Fiction is now available on YouTube 🙌🏼 I made a playlist for it with the seven chapters. Enjoy this time capsule from two decades ago!

SteeredMarigold: Steering Diffusion Towards Depth Completion of Largely Incomplete Depth Maps Even if the depth maps captured by RGB-D sensors deployed in real environments are often characterized by large areas missing valid depth measurements, the vast majority of depth completion methods st...

Great to see similar ideas to our own “SteeredMarigold” zero-shot depth completion method!

arxiv.org/abs/2409.10202

We were there too!

SteeredMarigold: Steering Diffusion Towards Depth Completion of Largely Incomplete Depth Maps Even if the depth maps captured by RGB-D sensors deployed in real environments are often characterized by large areas missing valid depth measurements, the vast majority of depth completion methods st...

Maybe this is it? arxiv.org/abs/2409.10202 @jakubgregorek.bsky.social