Gabriela Csurka retweeted
It's a marathon Sunday for @kgcs96 – co-author of PANst3R, Gabriela will also be giving 2 invited workshop talks AND the keynote dinner speech #WiCV - all the same day! 😅 @ICCVConference
CALIPOSE workshop: 🏃sites.google.com/view/calipo…
CRoCoDL workshop: 🏃♀️localizoo.com/workshop/
WICV: sites.google.com/view/wicv-i… 🏃♀️ 🧵6/9
Gabriela Csurka retweeted
We kick off #iccv25 Sun Oct 19th with PANSt3R - Multi-view Consistent Panoptic Segmentation
Paper: arxiv.org/abs/2506.21348
ILR+G workshop (morning): ilr-workshop.github.io/ICCVW…
OpenSUN workshop (afternoon): opensun3d.github.io/
TLDR: fast & scalable 3D reconstruction & scene understanding with object instances
@LojzeZust @_Anthea___ @bchidlovskii @JeromeRevaud @kgcs96
Yohann Cabon Leonid Antsfeld 🧵2/9
Gabriela Csurka retweeted
Gaussian Splatting Feature Fields for Privacy-Preserving Visual Localization
Maxime Pietrantoni, @kgcs96, @SattlerTorsten
tl;dr: align feature map and segmentation map with rendered feature map and segmentation map from Gaussian Splatting Feature Fields
arxiv.org/abs/2507.23569
Gabriela Csurka retweeted
I've been invited at the Weakly Supervised Computer Vision workshop at DeepLearning Indaba!
I'm super happy to present my works in front of 400 super motivated students!
Thanks to the organizers for inviting me! 😊 #DLIndaba2022 – at Tunisia
Gabriela Csurka retweeted
Introducing the latest model in @naverlabseurope *St3R family! In a single step PanSt3R performs fast & scalable 3D reconstruction & scene understanding with object instances.
Paper: arxiv.org/abs/2506.21348
More info on 3D FMs: tinyurl.com/vxw4mn9x
Accepted at #ICCV25 🎉
@LojzeZust @_Anthea___ @bchidlovskii @JeromeRevaud @kgcs96 Yohann Cabon Leonid Anstfeld
Gabriela Csurka retweeted
DUSt3R has come a long way since its release & is the foundation of hundreds of '3R' papers & new 3D vision models. This vid shows some of the models by the wonderful @naverlabseurope team (DUSt3R, MASt3R, MASt3R-sfm, MUSt3R) & there are more 🫢! ➡️europe.naverlabs.com/researc…
Congratulations to @JeromeRevaud @Vinc3nt_Leroy @bchidlovskii Yohann Cabon, @WeinzaepfelP Gabriela Csurka @kgcs96, @romain_bregier joined by @LojzeZust & fab collab with Shuze Wang, @BDuisterhof @LourdesAgapito @LStoffl @wbjang11
Big 🙏 to talented Ulysse Proton for integrating into the cool video 😊
Excited to share our latest work in the *St3R family. PanSt3R, accepted at #ICCV25, proposes a unified and integrated approach for 3D scene reconstruction and its panoptic segmentation in a single forward pass.
arxiv.org/abs/2506.21348
Gabriela Csurka retweeted
MUSt3R: Multi-view Network for Stereo 3D Reconstruction
TL;DR: DUSt3R guys strike back at CVPR '25!
Contributions:
• We revisit the DUSt3R architecture by making it symmetric and enabling N-view predictions in metric space. We entail it with a memory mechanism that decreases computational complexity for both offline and online reconstructions.
• We achieve state-of-the-art performance in unconstrained reconstruction scenarios, estimating field-of-view, camera pose, 3D reconstruction, and absolute scale without sacrificing inference speed.
Excited to share Maxime’s latest work on #GaussianSplatting Feature Fields for (#privacypreserving) #Visuallocalization. If interested Maxime will present his work tomorrow at #CVPR2025, Poster Session 1, Poster Number 85 (ExHall D).
Gabriela Csurka retweeted
Interestingly, the network generalizes very well to many more images than seen during training.
Code IS available here github.com/naver/must3r
I hope it works in your scenarios and you have as much fun as we do playing around with it!
Gabriela Csurka retweeted
Happy to introduce our new CVPR paper—Pow3R: Empowering Unconstrained 3D Reconstruction with Scene and Camera Priors : arxiv.org/pdf/2503.17316
Inspired by the amazing DUSt3R, which reconstructs 3D from just two unposed images, we explored:
(1/n)
Gabriela Csurka retweeted
Pizza break at @naverlabseurope before starting the last push to finish our #CVPR2025 papers!
Thanks @kgcs96 for having taken the photo 😃
Gabriela Csurka retweeted
We're nearly ready for the #HRISymposium panel on 'Design Principles for HRI' (1115CET) w/ Nazli Cila (@tudelft) Myeongsoo Shin (@NAVERLABSHQ) & @TitoFavaro (@PALRobotics) & host Jisun Park @naverlabseurope If #HRI #Design is your thing jump in ➡️ tinyurl.com/48sxxs3e !
Gabriela Csurka retweeted
MASt3R-SfM: a Fully-Integrated Solution for Unconstrained Structure-from-Motion
Paper: arxiv.org/abs/2409.19152
Code: github.com/naver/mast3r/tree…
Trying to break MASt3R-SfM. Not quite 😁
Code at github.com/naver/mast3r/tree… , scales to 1000+ images on a small GPU (also code for MASt3R+GLOMAP, see demo_glomap.py)
Gabriela Csurka retweeted
My first slide in my geometric vision class today with my two highlights of #ECCV24.
We presented the poster at the eXCV workshop at #ECCV2024 . Thank you to people stopping by. If you missed it and you are interested in explainability in CV (#xAI) here is our poster.
New paper on arXiv today. We study the problem of explaining with simple sentences the errors made by computer vision models - to "debug" their performance into the wild. Work led by Gabriela Csurka, with @TylerLHayes and @dlarlus (amazing team 🚀)
Paper: arxiv.org/abs/2408.04471
Gabriela Csurka retweeted
Publicity chair spotted, surrounded by the women of NLE - @CSProfKGD @kgcs96
Gabriela Csurka retweeted
Ladies and gentlemen, for the first time in history, you'll get to see the Test of Time award paper 10 years in advance. #DUSt3R
Joke aside, come pass say hi at #111, it's easy to remember!
If you'd like to fast track it to the DUSt3R poster it's #111 at Arch 4A-E ! Thanks @jon_barron
Gabriela Csurka retweeted
SPLATE: Sparse Late Interaction Retrieval
Adapts the ColBERTv2 model to map its embeddings to a sparse space, enabling efficient sparse retrieval for candidate generation in the late interaction paradigm.
📝arxiv.org/abs/2404.13950
Gabriela Csurka retweeted
Sparse late interaction retrieval. Fastest ColBERTv2 numbers ever.
I think this changes a lot of things, and I’m not surprised it comes from this team. At all.
SPLATE: Sparse Late Interaction Retrieval
Adapts the ColBERTv2 model to map its embeddings to a sparse space, enabling efficient sparse retrieval for candidate generation in the late interaction paradigm.
📝arxiv.org/abs/2404.13950