vannguyen

PhD student at @ImagineENPC | Previously intern at @RealityLabs and @SiemensMobility, visiting PhD student at @EPFL

[vannguyen @vannguyen_ng]

📢 Benchmark for 6D Object Pose Estimation 📢

BOP challenge 24 has been opened!

https://t.co/Od5U0LGzS9

Results to be presented at the R6D workshop at #ECCV2024

Details in comments below👇

[vannguyen @vannguyen_ng]

🚀 Open-sourced uAI-Nexus-MedVLM + MedVidBench (CVPR 2026)! A Medical Video LLM built for the OR, trained with our MedGRPO multi-task RL framework. Our 4B/7B models beat general-purpose GPT-5.4 & Gemini-3.1 on 8 medical video tasks.
Github: https://t.co/utyDVdrcap

[Tomáš Hodaň @tomhodan]

10th (!!) R6D workshop @ ICCV 2025: https://t.co/bFx9WxizlG

🤖 Object pose estimation for industrial robotics

📢 Strong speakers: @svlevine, @haoshu_fang, @MaxlDur , @GusKalra , @vaheta

📊 BOP Challenge 2025 with new BOP-Industrial datasets

With @ma_sundermeyer

[MrNeRF @janusch_patas]

Universal Beta Splatting

Contributions:
• Universal Beta Splatting: A unified N-dimensional representation with per-dimension shape control, enabling simultaneous modeling of spatial, angular, and temporal properties through anisotropic Beta kernels with spatial-orthogonal Cholesky parameterization.

• Efficient CUDA implementation: Fully accelerated differentiable rendering with custom kernels, achieving real-time performance.

• Interpretable scene decomposition: Emergent separation of geometric, appearance, and motion components through learned Beta parameters without supervision.

• Backward compatibility: Reduction to approximate existing methods as special cases, ensuring plug-in usability with performance lower bounds while enabling substantial improvements.

[Zhenjun Zhao @zhenjun_zhao]

Universal Beta Splatting

Rong Liu, @ZhongpaiGao, Benjamin Planche, @MeidaC, @vannguyen_ng, Meng Zheng, @AnwesaChoudhuri, Terrence Chen, @yuewang314, Andrew Feng, Ziyan Wu

tl;dr: radiance field rendering->N-dimensional anisotropic Beta kernels

https://t.co/HsBAkhjf7v

[Antoine Guédon @antoine_guedon]

⚠️Reconstructing sharp 3D meshes from a few unposed images is a hard and ambiguous problem.

☑️With MAtCha, we leverage a pretrained depth model to recover sharp meshes from sparse views including both foreground and background, within mins!🧵

🌐Webpage: https://t.co/di9e52XqFb

[Eric Brachmann @eric_brachmann]

Not long until the 9th(!) Workshop on Recovering 6D Object Pose (R6D) at #ECCV2024, Sunday AM.

Great speakers, and @vannguyen_ng, @tomhodan and @ma_sundermeyer will tell us about the #BOP Challenge 24 - the challenge is still running, but you get to see early bird results!

[Tomáš Hodaň @tomhodan]

📈 BOP update: Deadline for the 2024 challenge is extended to November 29!

BOP'24 focuses on model-free and model-based 2D/6D object detection, and introduces three new datasets from Meta and NVIDIA – HOT3D, HOPEv2, and HANDAL.

https://t.co/5nWv125raN

[Romain Loiseau, PhD @RomainLoiseau15]

#CVPR2024 poster presentation starting now at poster 358 🌟

[Romain Loiseau, PhD @RomainLoiseau15]

Starting now at poster 233 🌟

[Edward Miller @TweetEdMiller]

Introducing, HOT3D.

HOT3D is a new dataset from our team at Meta Reality Labs Research, to explore vision-based methods for hand-object interaction.

https://t.co/AR39UbhwgD

1/

[Tomáš Hodaň @tomhodan]

We are releasing 🔥HOT3D🔥, a new egocentric dataset for 3D hand and object tracking.

833 minutes (3.7M images) of multi-view image streams showing 19 subjects interacting with 33 objects, annotated with high-quality 3D poses of hands and objects.

Paper: https://t.co/9mKrH4aZKQ

[Imagine-ENPC @ImagineEnpc]

#CVPR2024 Thu 20 (AM)
GigaPose: Fast and Robust Novel Object Pose Estimation via One Correspondence
@vannguyen_ng @thibaultgroueix Mathieu Salzmann @VincentLepetit2
pdf: https://t.co/lP22K1uUkj
webpage: https://t.co/FoutSRgjE1
code: https://t.co/iK19JiIOYc

[Imagine-ENPC @ImagineEnpc]

#CVPR2024 Thu 20 (PM)
NOPE: Novel Object Pose Estimation from a Single Image
@vannguyen_ng @thibaultgroueix Georgy Ponimatkin,
  Yinlin Hu,  Renaud Marlet,  Mathieu Salzmann @VincentLepetit2
pdf: https://t.co/rMCU4zAccg
web: https://t.co/RElHEaXXDq
code: https://t.co/FQDemkDoM9

[Imagine-ENPC @ImagineEnpc]

#CVPR Fri 21 (AM)
OpenStreetView-5M: The Many Roads to Global Visual Geolocation
@g_astruc @nico_dufour @YSiglidis @Elt_Vincent @RomainLoiseau15 @xkungfu @vannguyen_ng @captnloic + others
pdf: https://t.co/0gwVnlbKSQ
web: https://t.co/47K5S6M7xV

[vannguyen @vannguyen_ng]

@ducha_aiki @TheZachMueller @wightmanr For example, this is how it looks like https://t.co/2ywe4c65uB

[vannguyen @vannguyen_ng]

@ducha_aiki @TheZachMueller @wightmanr I had exact same problem when uploading 3TB of BOP datasets to huggingface. Setting multi_commits=True solved the problem as it creates a PR with a stack of commits, when the connection is down, it automatically detects the last commits to avoid uploading from scratch

[Martin Sundermeyer @ma_sundermeyer]

The new BOP challenge 2024 just opened! 🔊 This year we are also competing on end-to-end *model-free* 6D object pose estimation! 🌟

After 5min/1GPU with an onboarding video of the target object, estimate the pose of the object in cluttered scenes.

[Eric Brachmann @eric_brachmann]

BOP is back with #ECCV2024. The presentation of the #BOP2023 results is just around the corner (at the CV4MR workshop, #CVPR2024), but we have no time to loose. Time for #BOP2024.

Thought pose estimation of unseen objects w/ a 3D model was hard? Try it based on a dynamic video!

[vannguyen @vannguyen_ng]

Besides, we also introduce three new datasets which enable evaluation of the model-based, and model-free tasks (all these datasets include both CAD models and onboarding videos): HOT3D captured from Aria glasses and Quest 3 from #Meta, HOPEv2 and HANDAL from #NVIDIA

[vannguyen @vannguyen_ng]

📢 Benchmark for 6D Object Pose Estimation 📢

BOP challenge 24 has been opened!

https://t.co/Od5U0LGzS9

Results to be presented at the R6D workshop at #ECCV2024

Details in comments below👇

[vannguyen @vannguyen_ng]

Dynamic onboarding: The object is manipulated by hands and the camera is either static (on a tripod) or dynamic (on a head-mounted device). Object masks for all video frames and the 6D object pose for the first frame are available.