Javier Civera

Robust Single-Parameter Point Cloud Registration using Truncated Lp Functions @SeongHunLee1, @patrick_vdwalle, @jcivera tl;dr: truncated L3 compatibility+support aggregation->correspondence ranking->hypotheses->GPU batch evaluation->L2/L0+ refinement arxiv.org/abs/2610.34271

🚀 Excited to share our new paper led by Zhisong and Fan (@YuishikiFan) F²SLAM Turning Feed-Forward Geometry into Persistent Factors for SLAM Feed-forward 3D predictions become measurements inside dense bundle adjustment, not a one-shot correction 📉 Uncalibrated Replica ATE: 0.002 m vs 0.030 m for the best feed-forward baseline 📄 arxiv.org/abs/2610.05207 💻 github.com/ZhisongXu/F2SLAM (code coming upon acceptance) 1/5

Kairos: Grounded Forecasting of Presence and Directional Flow in 4D Scene Graphs Iacopo Catalano, Julio A. Placed, @jcivera, Jorge Peña Queralta tl;dr: 3DSG->predictive directional-flow memory->4DSG arxiv.org/abs/2609.27467

VI3: Grounding Pretrained 3D Foundation Models with Inertial Cues Ernesto Lozano, Alberto Jaenal, @jcivera tl;dr: IMU + 3D foundation model (3DFM)->metric 3DFM arxiv.org/abs/2609.03824

ZipMVS: Multi-View Stereo with Compressed Cost Volumes Guanglin Jin, Hongshan Yu, @jcivera, Zhaoxin Li tl;dr: depth-sampling->adjust depth ranges; spatial cues->depth hypotheses arxiv.org/abs/2608.28033

🎉 New paper: SSMB A self-supervised keypoint detector that works directly on motion-blurred images No deblurring preprocessing, no handcrafted detector, no external pseudo-labels 📄 Paper: arxiv.org/abs/2608.27181 Under the hood: a Local Discriminability Enhancement (LDE) module restores fine local structure after global feature mixing, trained in two self-supervised stages: geometric pretraining on synthetic shapes, then blur-aware training on real sharp/blurred image pairs Proud to collaborate with @fabio_bellavia, Wenting Wang, @YuishikiFan, Jiajun Wu, Suryansh Kumar, Mingqiang Wei, Haoang Li, @jcivera Code, models and datasets coming upon acceptance #ComputerVision #KeypointDetection #SelfSupervisedLearning #ImageMatching

Visual Relocalization from Sparse Views in Aliased and Low-Texture Environments via Novel View Synthesis Maria Peribañez, @jcivera, Rudolph Triebel, @R_Giubilato tl;dr: MVSAnywhere-based depth & normal alignment loss + LiDAR-guided Chamfer-based loss arxiv.org/abs/2607.22147

SUFLECA: Scaling Up Feature Learning for CAD-to-image Alignment Saad Ejaz, Miguel Fernandez-Cortizas, @jcivera, Holger Voos, Jose Luis Sanchez-Lopez tl;dr: scaling + geometry-aware matching arxiv.org/abs/2607.15058

MyGO-Splat: Multi-Objective Closed-Loop Geometric Feedback for RGB-Only Gaussian SLAM @YuishikiFan, Ziyu Chen, @zhenjun_zhao, Zhisong Xu, Hui Zhu, Mingrui Li, Chunmao Jiang, @jcivera tl;dr: Gaussian map supervises tracking arxiv.org/abs/2606.29738

MAG-VLAQ: Multi-modal Aerial-Ground Query Aggregation for Cross-View Place Recognition Zhengyi Xu, Yuhang Ming, Zhihao Zhan, Hanyu Zhu, @jcivera, Wanzeng Kong tl;dr: DINOv2+Utonia->tokens; neural ODE-based RGB-LiDAR fusion+VLAQ->global descriptors arxiv.org/abs/2605.09418

PAGaS: Pixel-Aligned 1DoF Gaussian Splatting for Depth Refinement David Recasens, Robert Maier, @BozicAljaz, Stephane Grabli, @jcivera, Tony Tung, Edmond Boyer tl;dr: per-pixel spherical Gaussians for fitting back-projected pixel volumes arxiv.org/abs/2604.22129

See you at our ICLR 2026 poster! 🗓 Today, Apr 25 | 🕙 10:30 AM – 1:00 PM (BRT) 📌 Pavilion 3, #321 NPC — Neural Predictor-Corrector: Solving Homotopy Problems with Reinforcement Learning Come say hi!

Heading to ICLR 2026 in Rio de Janeiro 🇧🇷 Come find us at our poster on Saturday Apr 25, 10:30 AM – 1:00 PM (BRT), Pavilion 3, #321 If you want to chat about NPC, global optimization, homotopy methods, 3D vision, or anything else, feel free to say hi! Always happy to grab a coffee ☕ See you in Rio!