ECCV 2026 Opens in Malmö With Record 7,000 Attendees; Fei-Fei Li Team Wins Koenderink Prize

Conference chair Kalle Åström welcomed participants and said attendance exceeded 7,000, a record. Local exhibition service estimates put Malmö's hotel room supply at about 5,000, leaving a gap of more than 2,000 even if every hotel room were occupied, before accounting for residents, other tourists and business travelers. Many attendees crossed the Øresund Bridge from Copenhagen by train, and the article said a common greeting at the meeting was whether one had found a room in Malmö.

Program chair Atsuto Maki traced ECCV's Nordic history: Stockholm in 1994, Copenhagen in 2002 and now Malmö in 2026. He said Denmark and Sweden are close, Nordic cooperation is strong, and nearly 200 years of peace in the region have made such a gathering possible.

Program chair Anna Rohrbach presented paper trends. The conference received more than 10,000 submissions and accepted 2,834. Authors from Asia accounted for 65.4 percent, ahead of the United States at 18.5 percent and Europe at 14.2 percent. Image and video synthesis was the largest topic with more than 1,400 submissions, followed by vision, language and reasoning; 3D moved from its previous leading position to third. Rohrbach noted that the range of topics could not fit on one slide and required two.

The best paper award went to Heat Kernel Textures — the Geodesic Gaussians That Do Not Splat by Simone Foti, Caner Korkmaz, Stefanos Zafeiriou and Tolga Birdal, all from Imperial College London. The paper proposes Heat Kernel Textures, or HKTex, a texture representation based on discrete Riemannian geometry that uses anisotropic heat kernels as analogues of Gaussian functions in geodesic space. The committee called it a simple and novel contribution to surface texture representation and said it could have broad applications beyond the paper.

Best paper honorable mentions went to LSRM — High-fidelity Object-centric Reconstruction via Scaled Context Windows by Zhengqin Li and colleagues at Meta Reality Labs Research, and Poppy: Polarization-based Plug-and-play Guidance for Enhancing Surface Normal Estimation, mainly from the Computer Vision Lab at Stony Brook University. LSRM expands Transformer context windows to narrow the fidelity gap between feed-forward 3D object reconstruction and dense-view optimization, handling 20 times more object tokens and more than twice as many image tokens than prior best methods. Poppy is a training-free framework that uses polarization-encoded surface orientation to add physical cues for monocular surface normal estimators in reflection and textureless scenes.

The Koenderink Prize, which recognizes papers published at ECCV a decade ago with lasting impact, went to three works. One was Perceptual Losses for Real-time Style Transfer and Super-resolution from Fei-Fei Li's team at Stanford; the committee cited its introduction of a perceptual loss function based on pre-trained neural network features, now so common that its origin is almost forgotten. Another was SSD: Single Shot MultiBox Detector, by authors from Google, the University of North Carolina at Chapel Hill and the University of Michigan, Ann Arbor; the committee said it, along with R-CNN and YOLO, brought object detection into the deep learning era. The third was Learning without Forgetting by Zhizhong Li and Derek Hoiem; the paper addresses catastrophic forgetting in incremental learning through knowledge distillation and has been cited more than 5,900 times.

The PAMI Mark Everingham Community Contribution Award went to two teams: the FGVC (Fine-Grained Visual Categorization Challenge), with members from Caltech, Cornell, Edinburgh, Google and other institutions, held annually since 2015, and the VOT (Visual Object Tracking Challenge), jointly organized by the University of Birmingham, the University of Ljubljana, Czech Technical University in Prague and the Institute of Science and Technology Austria, which has updated and re-annotated tracking sequences annually since 2013.

ECVA awards were also presented. The 2026 ECVA Young Researcher Award went to Yuki Asano, a machine learning professor at the University of Technology Nuremberg who leads the FunAI Lab and earned his PhD at Oxford's Visual Geometry Group. The 2025 award went to Gül Varol, a permanent researcher in the IMAGINE team at ENPC in Paris. The ECVA PhD Award recognized Yao Fang, jointly trained at ETH Zurich and the Max Planck Institute for Intelligent Systems with advisers Michael J. Black and Marc Pollefeys and now a postdoc at Stanford, who could not attend due to a visa issue and was represented by Michael J. Black, and Paul-Edouard Sarlin, who earned his PhD at ETH Zurich and joined Google Zurich as a research scientist. The 2025 PhD award went to Yingjun Du, who earned his PhD at the University of Amsterdam and continued there as a postdoc.

The program includes three main conference days, 93 workshops, nine tutorials, three keynotes, two panels and multiple paper and poster sessions. Yann LeCun is scheduled to deliver a keynote. Jiajun Wu, an assistant professor at Stanford and a core researcher in world models and spatial intelligence, was among the busiest speakers, appearing in more than five workshops on topics including Physical Code, Agentic Robotics, Multisensory Reasoning and World Model Evaluation. Ping Tan, a professor at the Hong Kong University of Science and Technology and former head of Alibaba's XR Lab, spoke at the workshop Scalable 3D Scene Generation and 3D Geometric Scene Understanding about classical 3D geometry and scalable 3D scene reconstruction and generation.

Editor's Summary

ECCV 2026 opened in Malmö with record attendance of more than 7,000 and more than 10,000 submissions, with accepted authors concentrated in Asia. Its awards honored foundational computer vision work including perceptual loss, SSD and continual learning. The conference program featured 93 workshops and keynotes, reflecting the field's expansion into spatial intelligence, 3D and multimodal reasoning.