ZipSplat: Fewer Gaussians, Better Splats @veichta.bsky.social, Sunghwan Hong, @danielbarath.bsky.social, @marcpollefeys.bsky.social tl;dr: predict Gaussians directly in 3D from a compact set of scene tokens->decouple primitive placement from the 2D pixel grid arxiv.org/abs/2606.05102

SG2Loc: Sequential Visual Localization on 3D Scene Graphs Nicole Damblon, Olga Vysotska, Federico Tombari, @marcpollefeys.bsky.social, @danielbarath.bsky.social tl;dr: scene graph (SceneGraphLoc)->object meshes->semantic features->particle filtering arxiv.org/abs/2606.11880

DROID-SLAM in the Wild Moyang Li, Zihan Zhu, @marcpollefeys.bsky.social, @danielbarath.bsky.social tl;dr: uncertainty-aware BA->DROID-SLAM arxiv.org/abs/2603.19076

Global-Aware Edge Prioritization for Pose Graph Initialization @weitong8591.bsky.social @gtolias.bsky.social Jiri Matas, @danielbarath.bsky.social tl;dr: another global desc->GNN->MST. Supervision:# triangulated points/pair. +heuristic MST postprocessing. Eval on IMC-PT. arxiv.org/abs/2602.21963

Global-Aware Edge Prioritization for Pose Graph Initialization @weitong8591.bsky.social, @gtolias.bsky.social, Jiri Matas, @danielbarath.bsky.social tl;dr: rank pose graph edges->global consistency->improve SfM arxiv.org/abs/2602.21963

VIGS-SLAM: Visual Inertial Gaussian Splatting SLAM Zihan Zhu, Wei Zhang, Norbert Haala, @marcpollefeys.bsky.social, @danielbarath.bsky.social tl;dr: in title arxiv.org/abs/2512.02293

ReSplat: Learning Recurrent Gaussian Splats @haofeixu.bsky.social, @danielbarath.bsky.social, @andreasgeiger.bsky.social, @marcpollefeys.bsky.social tl;dr: rendering error->recurrent network->Gaussian updates arxiv.org/abs/2510.08575

SupeRANSAC: One RANSAC to Rule Them All @danielbarath.bsky.social tl;dr: most accurate, but slow-ish RANSAC, arxiv.org/abs/2506.04803 github.com/danini/super... Sampling: PROSAC + P-NAPSAC for homography & rigid pose, PROSAC for epipolar 1/

Excited to present our #CVPR2025 paper DepthSplat next week! DepthSplat is a feed-forward model that achieves high-quality Gaussian reconstruction and view synthesis in just 0.6 seconds. Looking forward to great conversations at the conference!

🏠 Introducing DepthSplat: a framework that connects Gaussian splatting with single- and multi-view depth estimation. This enables robust depth modeling and high-quality view synthesis with state-of-the-art results on ScanNet, RealEstate10K, and DL3DV. 🔗 haofeixu.github.io/depthsplat/

🏆 CrossOver is accepted as a 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁 at #CVPR2025! ✨ 💻 Fully open-sourced code with all pre-trained checkpoints: github.com/GradientSpac... 📡 Stay tuned for a deep-dive thread and what else we are cooking 🍳

Learning Affine Correspondences by Integrating Geometric Constraints Pengju Sun, Banglei Guan, Zhenbao Yu, Yang Shang, Qifeng Yu, @danielbarath.bsky.social tl;dr: DKM+AffNet? arxiv.org/abs/2504.04834

Joint work with @sayandsarkar.bsky.social, Ondrej Miksik, @marcpollefeys.bsky.social, and @danielbarath.bsky.social

🔗 arXiv: arxiv.org/abs/2502.15011 📂 Project page: sayands.github.io/crossover/ Joint work with Ondrej Miksik, @marcpollefeys.bsky.social, @danielbarath.bsky.social and @ir0armeni.bsky.social 🤝💡

Protip: Add poselib refinement to opencv magsac, you can thank me later.

Just wrapped up an incredible time at KEPAF conference with the Hungarian image processing, machine learning, and computer vision community — great talks, fresh ideas, and inspiring discussions. Excited to see you next time! #KEPAF #ComputerVision #MachineLearning