Lin Geng Foo

Computer Vision Postdoc at Max Planck Institute for Informatics

[VCAI - MPI for Informatics @VcaiMpi]

We’re excited to announce that the VCAI Department will co-organize three workshops at #CVPR2026! Details on the timeslots and location are provided in the thread below. We look forward to seeing you there!

The workshops, in chronological order, are:

[VCAI - MPI for Informatics @VcaiMpi]

Happy to announce that Christian Theobalt, Director of the VCAI Department at the MPI for Informatics, will deliver keynotes at several workshops at #CVPR2026.
If you’d like to meet him and learn more about our Institute, stop by and say hello!

Workshops in chronological order:

[Marc Habermann @marc_habermann]

Excited to share our work EgoRelight (SIGGRAPH 2026 Journal Track) ✨

We explore how to bring remote people into your physical surroundings as photorealistic, relightable avatars — instead of keeping everyone inside a fully virtual world.

[Lin Geng Foo @LinGengFoo]

@naiyanliziji Thanks! 😀

[Lin Geng Foo @LinGengFoo]

Excited to share our CVPR 2026 paper: Physical Simulator In-the-Loop Video Generation!

We introduce PSIVG, which brings a physical simulator into text-to-video inference, improving physical consistency in generated videos.

Project Page: https://t.co/cbL0HTNGEG

[Lin Geng Foo @LinGengFoo]

Very grateful to my collaborators @markkkhh , @alexlattas , @moschoglou , Thabo Beeler, and Christian Theobalt for this wonderful collaboration across MPI-INF, SUTD, A*STAR, and Google.

Code: https://t.co/dx9YEglHSt
Video: https://t.co/O94GQJivlW

[Lin Geng Foo @LinGengFoo]

Beyond guiding object dynamics such as gravity and collisions, we also introduce TTCO to better preserve object appearance during motion and rotation.

Please visit our poster at CVPR 2026 next week! We’d be happy to discuss the work in more detail!

[Wanyue Zhang @Wanyue__Z]

VHOI is accepted to CVPR 2026 Findings!

VHOI Enables controllable video generation for human-object interactions, turning easy-to-specify sparse trajectories into dense HOI-aware motion cues for realistic, instance-aware interaction videos.

🌐 Project: https://t.co/ayVwA9Ca4u

[Marc Habermann @marc_habermann]

🔥 #Eurographics2026
Can shoe insoles capture your full body motion? Step2Motion reconstructs diverse human locomotion from just pressure sensing insoles -- no cameras, no line-of-sight constraints needed!

[Eduardo Alvarado @roboonaut]

Have you ever wondered if we could capture human movement without mocap suits, VR trackers, or camera setups? What if all you needed was a pair of everyday insoles? 👟

Introducing Step2Motion, accepted to #Eurographics2026!

📝Project Page: https://t.co/JMxzexgc2M
🧵👇

[Marc Habermann @marc_habermann]

🔥 #3DV2026 Oral: We present a matrix-free Levenberg-Marquardt optimizer that makes second-order optimization practical for 3DGS! 🚀

[Keshigeyan Chandrasegaran @keshigeyan]

1/ Model architectures have been mostly treated as fixed post-training.

🌱 Introducing Grafting: A new way to edit pretrained diffusion transformers, allowing us to customize architectural designs on a small compute budget.

🌎 https://t.co/fjOTVqfVZr

Co-led with @MichaelPoli6

[MrNeRF @janusch_patas]

Second-order Optimization of Gaussian Splats with Importance Sampling

Contributions:
• Formulating Gaussian splatting as a non-linear least squares optimization problem, solved using a memory and computationally efficient Levenberg-Marquardt and conjugate gradient solver specifically tailored to 3DGS.

• Implementing a view and importance sampling strategy over the pixels (residuals) to effectively approximate the loss, significantly decreasing computational complexity.

• Developing an effective heuristic to determine the learning rate, eliminating the need for expensive line search methods while providing stable convergence for 3DGS optimization.

[Lin Geng Foo @LinGengFoo]

@NingDarlin59250 We are grateful for the interest in our work. Unfortunately, we are not planning to release the code for this work anytime soon. I have seen your email and have replied a few days ago. We'll surely let you know if there are any updates in the future.

[Lin Geng Foo @LinGengFoo]

I’m thrilled to share that I’ll be presenting two posters at #CVPR2024: “Action Detection via an Image Diffusion Process”, “LLMs are Good Sign Language Translators”!

I’ll be presenting during Thursday’s PM session. Poster IDs are: 362, 363. Please drop by if you’re attending!

[Lin Geng Foo @LinGengFoo]

Check out the papers here:
https://t.co/4lulTZd7n2
https://t.co/YqyBePVoIk

You can also check out our virtual presentations here:
https://t.co/DjnDatHKPj
https://t.co/yXBSsPVMPm

[Lin Geng Foo @LinGengFoo]

We thoroughly explore many single-modality settings (image, video, text, 3D, etc) and over 22 cross-modality settings.

We've carefully crafted it to be suitable for both seasoned researchers and those new to the field.

Paper: https://t.co/mrdAHSL4mH

#AI #AIGC #GenerativeAI

[Lin Geng Foo @LinGengFoo]

Excited to share our recent survey paper titled “AI-Generated Content (AIGC) for Various Data Modalities: A Survey”.

We comprehensively review AIGC methods across different data modalities, in both single-modality and cross-modality settings, citing 800+ references.

Overview:

[Lin Geng Foo @LinGengFoo]

If you’re at #ICCV2023 today, please visit poster at Nord – 097 for more information.
5 October 10:30 a.m. — 12.30 p.m. (Paris time)

You can also check out our virtual presentation here: https://t.co/k7gknZBDKZ

[Lin Geng Foo @LinGengFoo]

Excited to share our work at #ICCV2023: "Distribution-aligned Diffusion for Human Mesh Recovery"

Our paper outlines a diffusion-based method for 3D human mesh reconstruction from a single RGB image.

Paper: https://t.co/N5kDSEoORX