Senior researcher and principal investigator at MPI for Informatics working on 4D reconstruction, neural rendering and quantum-enhanced computer vision.
Saarbrücken, Germany
- Tweets 201
- Following 102
- Followers 971
- Likes 100
We are presenting today at #ECCV2026:
@GaurArihant et al. **NeuralGarSim: Geometry-Agnostic Garment Simulation with Neural Fields**
4–6 PM, ExHall # 115
* The same neural cloth simulator for different 3D representations
* Consistent simulation across different discretisations
5
52
2,590
On Tuesday, I gave a talk at the GeoInt Workshop at #ECCV2026 on "Inverse Geometric Inference with Neuroexplicit Models". Where should learning meet geometry? Here are a few ideas and projects that I'm happy to share and discuss: nextcloud.mpi-klsb.mpg.de/in….
geo-int.github.io/
22
1,534
Eager to learn how quantum computational paradigms are starting to shape computer vision research? Join us at the 3rd QCVML Workshop in Malmö, Sweden, on September 9 during #ECCV2026.
qcvml.github.io/
#QCVML2026 #量子コンピュータ #量子AI #quantumcomputing #QML
1
2
12
2,179
EgoForce: #SIGGRAPH2026 paper + demo in the Emerging Technologies Track
➣ Absolute 3D hand poses and shapes
➣ A single head-mounted camera
➣ Fisheye, perspective and wide-FOV sensors
➣ Up to 28% lower camera-space MPJPE on HOT3D
dfki-av.github.io/EgoForce/
#VR #AR #人工知能 #AI
5
38
4,205
New event-based 4D vision paper at #CVPR2026!
E-3DPSM rethinks real-time egocentric 3D human pose estimation as continuous state evolution, achieving
• 19% lower MPJPE and
• 2.7x better temporal stability
than previous methods.
4dqv.mpi-inf.mpg.de/E-3DPSM/
#ML #VR #人工知能 👇
1
3
37
3,350
E-3DPSM is a pose state machine that evolves latent states and predicts delta changes in 3D joint locations associated with the observed events. Deltas and direct 3D human pose predictions are fused using a neural Kalman-style filter, giving stable, drift-free 3D human poses.
👇
1
2
337
We gained several valuable insights for event-based 4D vision through this project, which was completed in a team with M. Deshmukh, @hiroyasu_akada, C. Theobalt and @HelgeRhodin.
arxiv.org/pdf/2604.08543
youtube.com/watch?v=bCfHkJyl…
github.com/MayurDeshmukh10/E…
#ML #VR #人工知能 #CVPR2026
1
5
368
Vlad Golyanik retweeted
I want to offer some unsolicited advice to computer vision researchers jumping into robotics. Don't focus too much on VLMs, VLAs etc. That's fine, but the real action is at the sensorimotor level. Most of the open problems in robotics are in manipulation, which is about hand-object interaction, and contacts and forces are central. Proprioception and tactile sensing are as important as vision. Don't get seduced by cherry-picked demos. You can't do robotics without doing robotics.
74
396
3,175
494,504
Humans do not navigate relying on accurate and complete 3D mental models of an environment; 2D cues often suffice. SceMoS adopts this principle, decoupling global intent from local scene-aware 3D motion generation.
🧍🤸🛏️🧘💻🧗
anindita127.github.io/SceMoS…
#ML #GenerativeAI #CVPR2026
3
18
2,008
ExposeAnyone, #CVPR2026 Findings, with @KaedeShioharaCS and @toshi_yamasaki:
• Deepfake detection via generative identity-consistency modelling
• Strong performance on #Sora2-generated videos in our new S2CFP dataset
arxiv.org/pdf/2601.02359
mapooon.github.io/ExposeAnyo…
#AI #CVPR 👇
2
3
20
2,505
Many modern supervised detectors will struggle to detect deepfakes from new generators.
In practice, the person’s identity is often known, and reference videos are available. We want to detect a deepfake of a certain person!🎙️
arxiv.org/pdf/2601.02359
#AI #CVPR👇[with sound]
1
1
391
✨Relightable Holoported Characters, #CVPR2026 (Oral & Award Candidate):
• Free-view rendering and relighting of full-body avatars
• A new light-stage capture strategy without OLATs
• Highly dynamic performances
vcai.mpi-inf.mpg.de/projects…
arxiv.org/pdf/2512.00255
#VR #CVPR
👇👇👇
2
13
95
7,487
• The core transformer-based module, RelightNet, efficiently approximates the rendering equation in a single forward pass
• Relighting becomes a learned rendering operator
• OLAT-style experiments indicate transferable reflectance properties under unseen illumination
👇
1
1
341
We thank our collaborators from Google. This work culminates a line of research on identity-specific full-body human avatar rendering and relighting that we started with Ashwath Shetty and @DiogoLuvizon.
Previous works:
vcai.mpi-inf.mpg.de/projects…
vcai.mpi-inf.mpg.de/projects…
2
151
Our first project leveraging @meta_aria glasses, led by Christen Millerdurai.
EgoForce estimates 3D hand poses from a single monocular egocentric camera in real time with high accuracy.
Done in collaboration with @DFKI.
youtu.be/hasL9g1k2aM
#AI #AR #VR #SIGGRAPH2026
💡 How do we build robust 3D hand tracking that generalizes across different wearable hardware?
The Augmented Vision team DFKI has taken a huge step toward solving this with EgoForce: Forearm-Guided Camera-Space 3D Hand Pose from a Monocular Egocentric Camera, just accepted at SIGGRAPH 2026.
Traditional monocular RGB methods suffer from depth-scale ambiguity and typically require laborious training on device-specific datasets. EgoForce bypasses this constraint by:
🌟 Leveraging a differentiable forearm representation to stabilize hand pose.
🫸 Utilizing a unified arm-hand transformer to predict geometry from a single egocentric view.
⚡️ Incorporating a ray-space closed-form solver for absolute 3D pose recovery.
Kudos to Alain Pagani, Christen Millerdurai, and the entire research team!👏
Website 👉dfki-av.github.io/EgoForce/
Paper 📰 dfki-av.github.io/EgoForce/s…
14
4,435
An important update to the arXiv Code of Conduct: "Incontrovertible evidence" of "the results of LLM generation" will lead to a one-year ban from arXiv. It is still unclear to what extent this new policy will be enforced, and we should raise awareness of it within our community.
Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/
2
392
Vlad Golyanik retweeted
✨#EG2026 STAR 🇩🇪
Non-rigid shape correspondenceサーベイ論文がドイツのアーヘンで開催されるEG STAR 2026にて発表されます。
基礎理論から最新の手法、課題、応用先まで初学者でも分かるよう丁寧にまとめてあります。
"Correspondence, Correspondence, Correspondence!" (金出武雄先生)
✨#Eurographics2026 STAR alert❗️
Our report on non-rigid 3D shape correspondences by A. Zhuravlev, L. Bastian et al. reviews foundations, recent methods & open challenges in this dynamically evolving research field. Join our talk in Aachen 🇩🇪 next month!
arxiv.org/pdf/2604.01274
3
5
1,598
🇧🇷 ∂∞-Grid by @NavamiK et al., #ICLR2026 🇧🇷
• Multi-resolution grids + RBF interpolation
• Infinitely differentiable
• Trained using differential equations (DEs) as losses
• Accurate modelling of physical fields
🔗4dqv.mpi-inf.mpg.de/DInf-Gri…
📝arxiv.org/pdf/2601.10715
[1/2]👇
1
17
130
12,887
𝐊𝐞𝐲 𝐢𝐧𝐢𝐭𝐢𝐚𝐥 𝐨𝐛𝐬𝐞𝐫𝐯𝐚𝐭𝐢𝐨𝐧: Most grid-based implicit representations rely on linear interpolation. They struggle with higher-order derivatives and, consequently, solutions to DEs. With @NavamiK, @shanthika_naik, @marc_habermann, A. Sharma and C. Theobalt.
[2/2]
2
349