Last reviewed
May 2, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 · 2,833 words
Add missing citations, update stale details, or suggest a clearer explanation.
The European Conference on Computer Vision (ECCV) is a biennial academic meeting on computer vision, held in even-numbered years at locations across Europe. Together with CVPR and ICCV it is one of the three top-tier conferences in the field, often referred to collectively as the "big three" of computer vision. ECCV was first held in Antibes, France in 1990, and has since visited cities in France, Italy, Sweden, the United Kingdom, Germany, Ireland, Denmark, the Czech Republic, Austria, Greece, Switzerland, the Netherlands, and Israel [1][2]. Since 2018 the conference has been organised by the European Computer Vision Association (ECVA), which also hosts an open-access proceedings archive at ecva.net [3].
ECCV is widely regarded as the European counterpart to ICCV and CVPR. It carries an A* rating in the Australian CORE 2026 ranking and an h5-index of 206 in Google Scholar Metrics, placing it among the most-cited venues in all of computer science [4][5]. The 2024 edition in Milan received 8,585 valid submissions and accepted 2,395 of them, an acceptance rate of 27.9% [6][7].
| Full name | European Conference on Computer Vision |
| Abbreviation | ECCV |
| Discipline | Computer vision, pattern recognition, machine learning |
| First held | April 1990, Antibes, France |
| Frequency | Biennial (even years) |
| Organiser | European Computer Vision Association (ECVA), since 2018 |
| Publisher | Springer (Lecture Notes in Computer Science); ECVA Open Access (since 2018) |
| CORE rank | A* (CORE 2026) |
| h5-index | 206 (Google Scholar Metrics) |
| Related conferences | CVPR, ICCV, BMVC, WACV |
The first ECCV was held at Antibes, on the French Riviera, from 23 to 27 April 1990. The proceedings were edited by Olivier Faugeras of INRIA and published by Springer as volume 427 of the Lecture Notes in Computer Science series [8]. The programme covered what were then the central problems of the field: image features, stereo and 3D reconstruction, optical flow, motion analysis, structure from motion, tracking, shape description, and recognition and matching. It was a relatively small gathering by modern standards, more comparable to a workshop than to today's multi-thousand-attendee event.
From that start, ECCV has rotated through European cities every two years without interruption. The early conferences moved among Santa Margherita Ligure (1992), Stockholm (1994), Cambridge in England (1996), Freiburg (1998), Dublin (2000), Copenhagen (2002), Prague (2004), Graz (2006), Marseille (2008) and Heraklion on Crete (2010). The pattern of holding the meeting in autumn became standard, with the exception of the very first edition.
Growth tracked the broader explosion of interest in computer vision and deep learning. ECCV 2010 in Crete drew about 1,174 submissions and accepted 325 papers, a rate of 27.6%, with only 38 talks selected for oral presentation [9]. By 2018 in Munich the submission count had jumped to 2,439 with 776 accepted [10]. ECCV 2020 (held virtually because of COVID-19) received 5,025 submissions and accepted 1,360 papers [11]. ECCV 2022 in Tel Aviv received 5,804 submissions, of which 1,645 were accepted (28.3%) [12]. ECCV 2024 in Milan reached 8,585 submissions, more than triple the 2018 figure and roughly seven times the 2010 figure [6].
The ECVA was founded in 2018 as a Swiss non-profit headquartered in Zurich, with the explicit purpose of organising the ECCV series and supporting future organising committees both scientifically and logistically [3]. Before ECVA, individual host institutions ran each edition, and the proceedings appeared exclusively through Springer LNCS. Since 2018 papers are also made available through the ECVA Open Access archive, mirroring the model used by the Computer Vision Foundation for CVPR and ICCV.
ECCV runs over five to six days. The standard layout has tutorials and workshops at the start and end of the week, with three days of the main technical programme in the middle. Accepted papers are split between a small number of oral presentations and a much larger pool of poster sessions. The oral track is highly selective: at ECCV 2010 only 3.2% of submissions were chosen for oral talks, while ECCV 2018 had 59 orals out of 776 accepted papers [9][10].
Review is double-blind, with each paper assigned to several reviewers and an Area Chair who coordinates the discussion and makes a recommendation. A rebuttal phase lets authors respond to reviews before the final decision. The programme committees for recent editions have ranged from several hundred to well over a thousand reviewers, drawn from academia and industry across the world. Acceptance rates have hovered around 25 to 30% across the modern era, similar to CVPR and ICCV.
Submission deadlines typically fall in early spring of the conference year, with notification in early summer and the final camera-ready in late summer. For the 2026 edition the paper registration deadline was 26 February 2026 and the submission deadline 5 March 2026, with decisions expected in mid-June and the conference itself running 8 to 12 September 2026 at the Malmö Arena and Malmömässan in Sweden [13].
| Edition | Year | Location | Country | Notes |
|---|---|---|---|---|
| 1 | 1990 | Antibes | France | Proceedings edited by Olivier Faugeras (INRIA) |
| 2 | 1992 | Santa Margherita Ligure | Italy | |
| 3 | 1994 | Stockholm | Sweden | |
| 4 | 1996 | Cambridge | United Kingdom | |
| 5 | 1998 | Freiburg | Germany | |
| 6 | 2000 | Dublin | Ireland | |
| 7 | 2002 | Copenhagen | Denmark | |
| 8 | 2004 | Prague | Czech Republic | |
| 9 | 2006 | Graz | Austria | |
| 10 | 2008 | Marseille | France | |
| 11 | 2010 | Heraklion, Crete | Greece | 1,174 submissions, 325 accepted (27.6%) |
| 12 | 2012 | Florence | Italy | |
| 13 | 2014 | Zürich | Switzerland | Microsoft COCO paper introduced |
| 14 | 2016 | Amsterdam | Netherlands | |
| 15 | 2018 | Munich | Germany | 2,439 submissions, 776 accepted; first edition organised under ECVA |
| 16 | 2020 | Glasgow | United Kingdom | Held virtually due to COVID-19; 5,025 submissions, 1,360 accepted |
| 17 | 2022 | Tel Aviv | Israel | 5,804 submissions, 1,645 accepted (28.3%) |
| 18 | 2024 | Milan | Italy | 8,585 submissions, 2,395 accepted (27.9%) |
| 19 | 2026 | Malmö | Sweden | 8 to 12 September 2026 at Malmö Arena and Malmömässan |
The ECCV Best Paper Award recognises the most outstanding contribution at each edition. The awards have tracked the major shifts in the field, from classical optimisation and geometry in the 2000s and early 2010s, through the deep learning era, into the current period dominated by self-supervised learning, large vision models, and 3D scene reconstruction.
| Year | Title | Authors |
|---|---|---|
| 2024 | Minimalist Vision with Freeform Pixels | Jeremy Klotz, Shree Nayar |
| 2022 | On the Versatile Uses of Partial Distance Correlation in Deep Learning | Xingjian Zhen, Zihang Meng, Rudrasis Chakraborty, Vikas Singh |
| 2020 | RAFT: Recurrent All-Pairs Field Transforms for Optical Flow | Zachary Teed, Jia Deng |
| 2018 | Implicit 3D Orientation Learning for 6D Object Detection from RGB Images | Martin Sundermeyer, Zoltan-Csaba Marton, Maximilian Durner, Manuel Brucker, Rudolph Triebel |
| 2016 | Real-Time 3D Reconstruction and 6-DoF Tracking with an Event Camera | Hanme Kim, Stefan Leutenegger, Andrew J. Davison |
| 2014 | Large-Scale Object Classification using Label Relation Graphs; Scene Chronology | Jia Deng et al.; Kyle Matzen, Noah Snavely |
| 2012 | Segmentation Propagation in ImageNet | Daniel Kuettel, Matthieu Guillaumin, Vittorio Ferrari |
| 2010 | Graph Cut based Inference with Co-occurrence Statistics | Lubor Ladicky, Chris Russell, Pushmeet Kohli, Philip H. S. Torr |
| 2008 | Learning Spatial Context: Using Stuff to Find Things | Geremy Heitz, Daphne Koller |
| 2006 | Learning to Combine Bottom-up and Top-down Segmentation | Anat Levin, Yair Weiss |
| 2004 | High Accuracy Optical Flow Estimation Based on a Theory for Warping | Thomas Brox, Andrés Bruhn, Nils Papenberg, Joachim Weickert |
A few of these papers had outsized influence on the field. RAFT (2020) became the standard reference architecture for optical flow estimation and was widely adopted in downstream video and 3D pipelines. The 2004 Brox et al. optical flow paper held up so well that it later won the Koenderink Prize itself, in 2014. The honourable mentions for 2024 included "Rasterized Edge Gradients" by Stanislav Pidhorskyi and colleagues at Meta, on differentiable rendering, and "Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Models" by Vitali Petsiuk and Kate Saenko on safety failures in diffusion model image generators [14].
The Koenderink Prize is a test-of-time award named after the Dutch vision scientist Jan Koenderink. It is awarded each ECCV to a paper presented ten years earlier at the conference whose impact has stood up over a decade of follow-up work. The prize has tracked many of the workhorse algorithms and datasets of computer vision.
| Year awarded | Original year | Paper |
|---|---|---|
| 2024 | 2014 | LSD-SLAM: Large-Scale Direct Monocular SLAM (Engel, Schöps, Cremers); Microsoft COCO: Common Objects in Context (Lin et al.) |
| 2022 | 2012 | A Naturalistic Open Source Movie for Optical Flow Evaluation (Sintel; Butler, Wulff, Stanley, Black); Indoor Segmentation and Support Inference from RGBD Images (NYU Depth v2; Silberman, Hoiem, Kohli, Fergus) |
| 2020 | 2010 | Improving the Fisher Kernel for Large-Scale Image Classification (Perronnin, Sánchez, Mensink); BRIEF: Binary Robust Independent Elementary Features (Calonder, Lepetit, Strecha, Fua) |
| 2018 | 2008 | Hamming Embedding and Weak Geometric Consistency for Large Scale Image Search (Jégou, Douze, Schmid); Semi-supervised On-Line Boosting for Robust Tracking (Grabner, Leistner, Bischof) |
| 2016 | 2006 | SURF: Speeded Up Robust Features (Bay, Tuytelaars, Van Gool); Machine Learning for High-Speed Corner Detection (FAST; Rosten, Drummond) |
| 2014 | 2004 | Face Recognition with Local Binary Patterns (Ahonen, Hadid, Pietikäinen); High Accuracy Optical Flow Estimation Based on a Theory for Warping (Brox, Bruhn, Papenberg, Weickert) |
| 2012 | 2002 | What Energy Functions Can Be Minimized via Graph Cuts? (Kolmogorov, Zabih) |
| 2010 | 2000 | Stochastic Tracking of 3D Human Figures Using 2D Image Motion (Sidenbladh, Black, Fleet); Unsupervised Learning of Models for Recognition (Weber, Welling, Perona) |
Several of these papers are now standard fare in any computer vision textbook. SURF and BRIEF underpin a generation of feature-matching pipelines used in everything from panorama stitching to robot localisation. The Microsoft COCO dataset became the de facto benchmark for object detection, instance segmentation, keypoint detection, and image captioning, and it was a key training resource for the early wave of large vision models. LSD-SLAM was a landmark direct (rather than feature-based) SLAM system that helped popularise dense monocular reconstruction.
The ECVA also administers a set of associated prizes that are presented at ECCV.
Workshops bracket the main conference at the start and end of the week. They are themed around emerging topics or specific application areas, and they often include their own paper tracks, invited talks, and competitions. Recent editions have hosted dozens of workshops on subjects ranging from autonomous driving and medical imaging to vision-language models, video generation, ethical AI, and 3D scene understanding. ECCV 2022 in Tel Aviv ran on the order of 60 workshops and tutorials, and ECCV 2024 in Milan ran a similar count, with workshop proceedings published as additional volumes in the Springer LNCS series [15].
Tutorials run on the same outer days and tend to provide deeper introductions to specific subfields, often presented by leading researchers in those areas. Topics in recent years have included neural radiance fields, transformers for vision, self-supervised learning, and large multimodal models. A demo and exhibitor programme runs alongside the main sessions, with hardware vendors, cloud providers, and AI labs presenting tools and recruiting researchers.
Many of the techniques that define modern computer vision had their first or definitive presentation at ECCV. The 2004 Brox et al. optical flow paper anchored variational methods for years; SURF (2006) and BRIEF (2010) became default feature detectors and descriptors; the original Microsoft COCO dataset (2014) became one of the most influential benchmark datasets in machine learning; and RAFT (2020) reset the state of the art on optical flow. The Sintel optical flow benchmark (2012) and NYU Depth v2 (2012) likewise became reference datasets used in thousands of follow-up papers.
ECCV has also been a venue for the steady accumulation of less spectacular but vital advances: improved network architectures, training tricks, evaluation methodologies, and dataset analyses that collectively shape how the field works. Because it sits one year off from the CVPR-ICCV cycle (CVPR is annual, ICCV is biennial in odd years, ECCV is biennial in even years), the European meeting often serves as a checkpoint where ideas first floated at CVPR or ICCV are refined and extended.
The three top computer vision conferences differ mostly in geography, sponsorship, and cadence rather than in scope or scientific standing.
| Aspect | ECCV | CVPR | ICCV |
|---|---|---|---|
| Frequency | Biennial (even years) | Annual | Biennial (odd years) |
| Region | Europe | Mostly North America | Rotating worldwide |
| Sponsor | ECVA | IEEE Computer Society and CVF | IEEE Computer Society and CVF |
| First held | 1990 (Antibes) | 1983 (Washington, D.C.) | 1987 (London) |
| Proceedings | Springer LNCS; ECVA Open Access | CVF Open Access; IEEE Xplore | CVF Open Access; IEEE Xplore |
| Recent submissions | 8,585 (2024) | 13,008 (2025) | over 8,000 (2023) |
In practice researchers treat publications at all three venues as equivalent in prestige. A typical computer vision PhD will submit to whichever of the three has the next deadline, and many groups time their projects around the rotating cycle. Workshops and tutorials cross-pollinate between the three meetings, and the same Area Chairs and senior reviewers often serve at all of them. The CORE 2026 conference rankings give all three an A* rating in the field of computer vision and multimedia computation [4].
For most of its history ECCV proceedings appeared exclusively through Springer as part of the Lecture Notes in Computer Science (LNCS) series. The first volume was LNCS 427 (1990), and recent editions have spanned dozens of volumes per conference because of the sheer number of accepted papers. Springer continues to publish the official proceedings, and individual papers also appear as Springer chapters with DOIs.
Since 2018 the ECVA has additionally maintained an open-access mirror at ecva.net, where all main-conference papers and most workshop papers can be downloaded freely. This dual model gives ECCV a similar open-access posture to CVPR and ICCV, whose papers are made freely available through the Computer Vision Foundation Open Access archive. Older ECCV proceedings (pre-2018) are still primarily accessed through Springer Link, although many individual papers are also available on author websites and on arXiv.