Alexander Veicht

PhD Student @cvg_ethz, ETH Zurich | 3D Computer Vision | Machine Learning

[Alexander Veicht @veichta]

Excited to release ZipSplat: Fewer Gaussians, Better Splats! 🎉

Feed-forward 3D Gaussian Splatting that predicts Gaussians directly in 3D with fewer Gaussians and better splats. Pose-free, in under a second.

👉 Try the live demo in your browser: https://t.co/kH3J0rfHVm

🧵👇

[Haofei Xu @haofeixu]

Does 3D reconstruction have to be complex?

We answer this question with PointDiT (#ICML2026): a minimalist pixel-space Diffusion Transformer without bells and whistles.

We show that a plain ViT can estimate dense 3D point maps by operating directly on raw patches. No hybrid ViT+Conv architectures, no lossy VAEs, no complicated training losses. (1/5) 🧵👇

[Xi @xiwang1212]

Happy to share that I’ll be joining Ulm University this fall! We have open PhD positions in egocentric vision and embodied AI, with opportunities for co-supervision by Dima Damen @dimadamen and close collaboration with the Aria team. More details below 👇

[Roni Itkin @ItkRoni]

We introduce 🌍GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens.🌍

Most feed-forward 3DGS methods still start from pixel, voxel, or dense view-aligned primitives.
We take a different route: align first, decode later. 🧵👇

[Alexander Veicht @veichta]

@zhenjun_zhao @majti89 @mapo1 Thanks for sharing! 🙏 Feel free to try the interactive demo on the project page: https://t.co/kH3J0rfHVm

[Alexander Veicht @veichta]

7/🕹️ Try it yourself: Check out the interactive demo on the project page, or spin up the demo in the repo to reconstruct your own scenes!

🌐 Project page: https://t.co/kH3J0rfHVm
💻 Code: https://t.co/l1mgg1ScgT

Huge thanks to my supervisors Sunghwan Hong, @majti89 & @mapo1 🙏

[Alexander Veicht @veichta]

Excited to release ZipSplat: Fewer Gaussians, Better Splats! 🎉

Feed-forward 3D Gaussian Splatting that predicts Gaussians directly in 3D with fewer Gaussians and better splats. Pose-free, in under a second.

👉 Try the live demo in your browser: https://t.co/kH3J0rfHVm

🧵👇

[Alexander Veicht @veichta]

6/ 💭 The takeaway: most methods uplift 2D predictions along rays, tying each gaussian to the estimated camera and depth, so errors displace the geometry. ZipSplat predicts directly in 3D for a more 3D-consistent representation with fewer Gaussians and better splats.

[Viktor Larsson @visionviktor]

Interested in doing research in 3D computer vision, geometry and machine learning? Open PhD position available in my group. Check out https://t.co/NV0sJqsz35 for more details on how to apply. Application deadline June 17th! Feel free to contact me if you have any questions.

[Dmytro Mishkin 🇺🇦 @ducha_aiki]

Colmap 3.11 release,a LOT of of cool things:
- incremental mapper with absolute pose priors
- CUDA-based BA through Ceres (disabled by default).
- PoseLib's RANSACs
- New BA covariance estimation,faster & more robust than Ceres.
- Fix Affine-covariant SIFT
https://t.co/FrRnMcuSgN

[Paul-Edouard Sarlin @pesarlin]

Introducing GeoCalib for camera calibration: gravity & intrinsics from a single image 📸
➡️Differentiable geometry optimization FTW!
➡️Video https://t.co/kecW4fg5fr
➡️Demo https://t.co/e0RawJn22e
➡️Paper https://t.co/GQr1KoosMf
by @veichta with @PhilippCSE @mapo1 for #ECCV2024 1/

[Alexander Veicht @veichta]

@Parskatt @abcamiletto @Hydra Yes since hydra is based on omegaconf so they work very well together. I’ve started with omegaconf only so some command line parsing is a bit cluttered but otherwise it works very well.

[Alexander Veicht @veichta]

@Parskatt @abcamiletto @Hydra I’ve used hydra in GeoCalib and was quite happy with it :) it’s not perfect but if you want everything in a single place, it is not a bad option.

[Zhenjun Zhao @zhenjun_zhao]

GeoCalib: Learning Single-image Calibration with Geometric Optimization

@veichta, @pesarlin, @PhilippCSE, @mapo1

tl;dr: input image->DNN (SegNeXt based)->pixel-wise Perspective Field+confidences->iterative LM optimization->train end-to-end DNN

https://t.co/QafYHI60mi

[Sonia Health @Sonia_Health]

We’re excited to introduce Sonia, an AI cognitive therapist who is available anytime, anywhere and for anyone 🎉🧠🌍

Sonia is available for free on the App Store - Download and start your first session today! https://t.co/bz5aBC1Im4

[Kaggle @kaggle]

👏🏼Congratulations to all of the winners who presented their solutions for the @GoogleAI Image Matching Challenge 2023 at @CVPR last month. Check out the recording here (Kaggle talks start at 3:37:00): https://t.co/rAbTRpwuvE

[Paul-Edouard Sarlin @pesarlin]

You liked SuperGlue? You'll love ⚡️LightGlue⚡️, our new deep network for light-speed image matching!
➡️Faster, stronger, easier to train than SuperGlue
➡️Code: https://t.co/SLLvKOwbEo
➡️Paper: https://t.co/8b38JCoFdg
Fantastic work by @PhilippCSE for #ICCV2023, with @mapo1
1/