Eshed Ohn-Bar

ZeroVO: Visual Odometry with Minimal Assumptions A novel visual odometry algorithm that achieves zero-shot generalization across diverse cameras and environments.

State-of-the-art performance across datasets (KITTI, nuScenes, Argoverse, and a newly introduced GTA dataset)

Project page: zvocvpr.github.io

1) We show a geometry-guided, cross-attention architecture enables robust VO generalization, including under noisy estimated camera parameters.

(2) We develop a novel multimodal semi-supervised training framework that facilitates further generalization.

Come by our poster #122 tomorrow 6/14 PM session at @cvprconference.bsky.social #CVPR #CVPR2025

ZeroVO is a new algorithm for generalized, cross-settings metric-scale visual odometry (VO)

Two key insights:

JP Williams – Blind singer, songwriter, and accessibility tech professional

And yes… there will be live music (we are in Nashville, after all)

accessibility-cv.github.io

Kristen Grauman – UT Austin - Video Understanding for Accessible Skill Learning

Gül Varol – École des Ponts ParisTech - Towards Open-Vocabulary Sign Language Translation

Bolei Zhou – UCLA - AI-assisted Micro-mobility through Large-scale Real2Sim Simulation

👇

Join us for the AVA Workshop at CVPR 2025 June 11 (afternoon) - with incredible speakers working at the intersection of computer vision, systems, and assistive technologies.
👇

Davide Scaramuzza – University of Zurich - Visual Assistance for Blind People

#CVPR #CVPR2025 @cvprconference.bsky.social

I do hope more companies will join - we really could use more participants... But hard for me to see that happening currently - I think we may need one more research cycle of generalist AI agents, then things can democratize a bit and companies can productize more easily.

⌛𝙏𝙬𝙤. 𝙢𝙤𝙧𝙚. 𝙬𝙚𝙚𝙠𝙨.

to put your human and language models to the test.

🧩We’ve designed a unique challenge in assistive motion and instruction generation at @cvprconference.bsky.social - excited to see what you can do.

Winning submissions get $500 #CVPR #CVPR2025 @dr-g.bsky.social

Maybe LLM response?

Lidar is great for self-supervision, but tesla can get very similar data (with pseudo-lidar/dense monodepth maturing), and using offline video with past/future temporal consistency to resolve remaining ambiguities.

To do my part in saving energy, I never say 'please' or 'thank you' to chatGPT.

Excited for the Accessibility, Vision, and Autonomy workshop at CVPR 2025

Participate in 3D motion and instruction generation challenges to win prizes and advance assistive system

🌐 More details: lnkd.in/dr6BQrPE
🦾 Challenge: June 8th

@cvprconference.bsky.social #CVPR2025 #CVPR

Unless openreview crashes 😀

Amazing work - continual learning could help scale further infdriver.github.io

Sub 0C temperatures?

We go golfing

Bluesky seems nice. I would also have supported Myspace

Musicals are definitely worst on the list😀