Last reviewed
May 2, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 3,595 words
Add missing citations, update stale details, or suggest a clearer explanation.
The International Conference on Computer Vision (ICCV) is a biennial academic conference held in odd-numbered years and devoted to research in computer vision. It is sponsored by the IEEE Computer Society's Pattern Analysis and Machine Intelligence (PAMI) Technical Committee, and since 2013 it has been co-sponsored by the Computer Vision Foundation. Together with CVPR (annual) and ECCV (biennial in even years), ICCV is one of the "big three" conferences in computer vision and one of the most selective venues in the field. The conference's best-paper award, the Marr Prize, is named after British neuroscientist David Marr and has gone to several of the most cited papers in modern computer vision, including Mask R-CNN (2017) and the Swin Transformer (2021).
| Field | Value |
|---|---|
| Full name | International Conference on Computer Vision |
| Abbreviation | ICCV |
| Discipline | Computer vision |
| Frequency | Biennial (odd-numbered years) |
| First held | June 8 to 11, 1987, London, United Kingdom |
| Sponsor | IEEE Computer Society (PAMI Technical Committee), Computer Vision Foundation |
| Best paper award | Marr Prize (named for David Marr) |
| Related conferences | CVPR, ECCV, NeurIPS |
| CORE rank | A* |
| Most recent edition | ICCV 2025, Honolulu, Hawai'i (October 19 to 23, 2025) |
| Website | iccv.thecvf.com |
The first ICCV was held at the Royal National Hotel in London from June 8 to 11, 1987, sponsored by the IEEE Computer Society in cooperation with the International Association for Pattern Recognition. The inaugural meeting drew roughly 300 attendees and accepted about 60 papers, with topics centred on what would later be called classical computer vision: edge detection, segmentation, stereo, shape from shading, and early 3D reconstruction. David Heeger took home the very first Marr Prize for "Optical Flow using Spatiotemporal Filters," and Michael Kass, Andrew Witkin, and Demetri Terzopoulos received an honourable mention for "Snakes: Active Contour Models," a paper that would go on to be cited tens of thousands of times.
The second edition followed quickly in Tampa, Florida in December 1988, but starting with the third edition in Osaka in 1990 the meeting settled into a roughly biennial pattern. From 1999 onward, ICCV has been held every two years in odd-numbered years, alternating with ECCV in even years so that the European and IEEE-led conferences do not collide. (CVPR, the third major venue, runs every year, usually in June.) Locations rotate across continents to share the cost and effort of hosting and to keep the meeting genuinely international.
The conference grew slowly at first, then very quickly after 2012, when deep learning broke into the field via AlexNet at NeurIPS. ICCV submissions roughly doubled between 2017 and 2019, then doubled again between 2019 and 2025. By the 2025 edition in Honolulu the program committee was processing more than 11,000 submissions and accepting around 2,700.
ICCV uses a double-blind peer review process. Authors submit anonymized PDFs in March, reviewers and area chairs work through them in the spring, and decisions are released in early summer. The conference itself spans roughly a week. The first day or two is typically devoted to tutorials and workshops; the main technical program then runs for three or four days, with a mix of single-track oral sessions for highlighted papers, very large poster sessions for the rest, plenary keynotes, an industrial expo, and an awards ceremony. Workshops and tutorials resume on the final day or two.
Acceptance rates have stayed in a fairly tight band of 24 to 31 percent across the past decade, even as raw submission volume has grown by nearly an order of magnitude. The table below uses figures collected by Paper Copilot from the ICCV organisers' opening slides.
| Edition | Submissions | Accepted | Acceptance rate |
|---|---|---|---|
| ICCV 2011 (Barcelona) | 1,433 | 339 | 23.7% |
| ICCV 2013 (Sydney) | 1,629 | 454 | 27.9% |
| ICCV 2015 (Santiago) | 1,698 | 525 | 30.9% |
| ICCV 2017 (Venice) | 2,143 | 621 | 29.0% |
| ICCV 2019 (Seoul) | 4,303 | 1,062 | 24.7% |
| ICCV 2021 (virtual) | 6,152 | 1,612 | 26.2% |
| ICCV 2023 (Paris) | 8,620 | 2,155 | 25.0% |
| ICCV 2025 (Honolulu) | 11,152 | 2,698 | 24.2% |
A small fraction of accepted papers are picked for oral presentation; the rest appear as posters and have ten or twelve minutes of recorded video plus a long poster session. All accepted papers are published in the IEEE Computer Society Digital Library and, since 2013, also in the open-access CVF Open Access repository, where they can be downloaded for free.
| Year | Edition | Location | Notes |
|---|---|---|---|
| 1987 | 1st | London, United Kingdom | Royal National Hotel, June 8 to 11 |
| 1988 | 2nd | Tampa, Florida, United States | December |
| 1990 | 3rd | Osaka, Japan | First time in Asia |
| 1993 | 4th | Berlin, Germany | First time in continental Europe |
| 1995 | 5th | Cambridge, Massachusetts, United States | |
| 1998 | 6th | Mumbai (Bombay), India | First time in South Asia |
| 1999 | 7th | Kerkyra (Corfu), Greece | Start of strict biennial odd-year cadence |
| 2001 | 8th | Vancouver, Canada | |
| 2003 | 9th | Nice, France | |
| 2005 | 10th | Beijing, China | |
| 2007 | 11th | Rio de Janeiro, Brazil | First time in South America |
| 2009 | 12th | Kyoto, Japan | |
| 2011 | 13th | Barcelona, Spain | November 6 to 13 |
| 2013 | 14th | Sydney, Australia | First time in Oceania |
| 2015 | 15th | Santiago, Chile | December |
| 2017 | 16th | Venice, Italy | Lido island, October 22 to 29 |
| 2019 | 17th | Seoul, South Korea | COEX Convention Center, October 27 to November 2 |
| 2021 | 18th | Virtual (originally Montreal, Canada) | October 11 to 17, moved online due to COVID-19 |
| 2023 | 19th | Paris, France | Paris Convention Center, October 2 to 6 |
| 2025 | 20th | Honolulu, Hawai'i, United States | Hawai'i Convention Center, October 19 to 23 |
The pattern of locations gives a quick history of where the centre of gravity in computer vision research has sat. The 1980s and early 1990s editions were almost evenly split between North America, Europe, and Japan; from 2005 onward China, Korea, Brazil, Australia, and Chile have all hosted at least once, reflecting the global spread of vision research groups.
ICCV's best paper award is the Marr Prize, named after David Marr, the British neuroscientist whose 1982 book Vision helped establish the modern computational view of perception. The prize has been awarded at every ICCV since 1987, sometimes to a single paper and sometimes split among two or three. Several Marr Prize papers have gone on to become foundational references in their subfield. The list below collates winners from the IEEE Computer Society PAMI committee, the Computer Vision Foundation awards page, and the ICCV organisers' opening sessions.
| Year | Paper | Authors |
|---|---|---|
| 1987 | Optical Flow using Spatiotemporal Filters | David Heeger |
| 1988 | Color from Black and White | Brian Funt, Jian Ho |
| 1990 | Shape from Interreflections | Shree Nayar, Katsushi Ikeuchi, Takeo Kanade |
| 1993 | Extracting Projective Structure from Single Perspective Views of 3D Point Sets | Charles A. Rothwell, David A. Forsyth, Andrew Zisserman, Joseph L. Mundy |
| 1995 | A Theory of Specular Surface Geometry | Michael Oren, Shree Nayar |
| 1995 | Shape from Shading with Interreflections under a Proximal Light Source | Toshikazu Wada, Hiroyuki Ukida, Takashi Matsuyama |
| 1998 | Self-Calibration and Metric Reconstruction in spite of Varying and Unknown Internal Camera Parameters | Marc Pollefeys, Reinhard Koch, Luc Van Gool |
| 1998 | The Problem of Degeneracy in Structure and Motion Recovery from Uncalibrated Image Sequences | Phil Torr, Andrew Fitzgibbon, Andrew Zisserman |
| 1999 | A Theory of Shape by Space Carving | Kiriakos Kutulakos, Steven Seitz |
| 1999 | Euclidean Reconstruction and Reprojection up to Subgroups | Yi Ma, Stefano Soatto, Jana Kosecka, Shankar Sastry |
| 2001 | Probabilistic Tracking in a Metric Space | Kentaro Toyama, Andrew Blake |
| 2001 | The Space of All Stereo Images | Steven Seitz |
| 2003 | Image-Based Rendering using Image-Based Priors | Andrew Fitzgibbon, Yonatan Wexler, Andrew Zisserman |
| 2003 | Image Parsing: Unifying Segmentation, Detection and Recognition | Zhuowen Tu, Xiangrong Chen, Alan L. Yuille, Song-Chun Zhu |
| 2003 | Detecting Pedestrians using Patterns of Motion and Appearance | Paul Viola, Michael J. Jones, Daniel Snow |
| 2005 | Globally Optimal Estimates for Geometric Reconstruction Problems | Fredrik Kahl, Didier Henrion |
| 2007 | Population Shape Regression From Random Design Data | Bradley Davis, P. Thomas Fletcher, Elizabeth Bullitt, Sarang Joshi |
| 2009 | Discriminative Models for Multi-Class Object Layout | Chaitanya Desai, Deva Ramanan, Charless Fowlkes |
| 2011 | Relative Attributes | Devi Parikh, Kristen Grauman |
| 2013 | From Large Scale Image Categorization to Entry-Level Categories | Vicente Ordonez, Jia Deng, Yejin Choi, Alexander Berg, Tamara Berg |
| 2015 | Deep Neural Decision Forests | Peter Kontschieder, Madalina Fiterau, Antonio Criminisi, Samuel Rota Bulo |
| 2017 | Mask R-CNN | Kaiming He, Georgia Gkioxari, Piotr Dollar, Ross Girshick |
| 2019 | SinGAN: Learning a Generative Model from a Single Natural Image | Tamar Rott Shaham, Tali Dekel, Tomer Michaeli |
| 2021 | Swin Transformer: Hierarchical Vision Transformer using Shifted Windows | Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo |
| 2023 | Passive Ultra-Wideband Single-Photon Imaging | Mian Wei, Sotiris Nousias, Rahul Gulve, David B. Lindell, Kiriakos N. Kutulakos |
| 2023 | Adding Conditional Control to Text-to-Image Diffusion Models (ControlNet) | Lvmin Zhang, Anyi Rao, Maneesh Agrawala |
| 2025 | BrickGPT: Generating Physically Stable and Buildable Brick Structures from Text | Ava Pun, Kangle Deng, Ruixuan Liu, Deva Ramanan, Changliu Liu, Jun-Yan Zhu |
A few entries are worth pulling out. The 1987 prize to Heeger came alongside the famous "Snakes" honourable mention, which introduced active contour models that are still used in medical imaging textbooks. The 1990 prize to Nayar, Ikeuchi, and Kanade established interreflection modelling as a serious topic. The 1998 prize to Pollefeys, Koch, and Van Gool helped kick off practical self-calibration in structure-from-motion. Mask R-CNN in 2017 has, by any reasonable count, been the most influential single ICCV paper of the deep-learning era; it became the default architecture for instance segmentation almost overnight and now sits at well over a hundred thousand citations. Swin Transformer in 2021 played a similar role in pushing transformer architectures into mainstream computer vision after vision transformer work showed it was possible. ControlNet in 2023 was an unusual award in that the underlying paper had already become a standard tool in the Stable Diffusion community before it was even formally presented at the conference.
ICCV typically gives out two to four honourable mentions alongside the Marr Prize. A short list of recent ones, drawn from the official CVF awards page:
| Year | Paper | Authors |
|---|---|---|
| 2007 | Deformable Template As Active Basis | Y. Nian Wu, Z. Si, C. Fleming, S.-C. Zhu |
| 2009 | Looking Around the Corner Using Transient Imaging | A. Kirmani, T. Hutchison, J. Davis, R. Raskar |
| 2013 | Piecewise Rigid Scene Flow | C. Vogel, K. Schindler, S. Roth |
| 2015 | Holistically-Nested Edge Detection | S. Xie, Z. Tu |
| 2017 | Open Set Domain Adaptation | P. P. Busto, J. Gall |
| 2019 | Asynchronous Single-Photon 3D Imaging | A. Gupta, A. Ingle, M. Gupta |
| 2021 | Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields | J. T. Barron, B. Mildenhall, M. Tancik, P. Hedman, R. Martin-Brualla, P. Srinivasan |
| 2023 | Segment Anything | A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollar, R. Girshick |
| 2025 | RayZer: A Self-supervised Large View Synthesis Model | H. Jiang, H. Tan, P. Wang, H. Jin, Y. Zhao, S. Bi, K. Zhang, F. Luan, K. Sunkavalli, Q. Huang, G. Pavlakos |
Segment Anything in 2023 is a notable case: the paper, from Meta AI, was already a household name in the foundation model community by the time of the conference. It received an honourable mention rather than the Marr Prize itself, with the prize going to two more research-y papers (one on novel imaging hardware and one on diffusion model control).
A separate Best Student Paper Award goes to a paper whose lead author is a student. Recent winners include:
| Year | Paper | Lead student author |
|---|---|---|
| 2011 | Close the Loop: Joint Blind Image Restoration and Recognition with Sparse Representation Prior | Haichao Zhang |
| 2017 | Focal Loss for Dense Object Detection | Tsung-Yi Lin |
| 2019 | PLMP: Point-Line Minimal Problems in Complete Multi-View Visibility | Timothy Duff |
| 2021 | Pixel-Perfect Structure-from-Motion with Featuremetric Refinement | Philipp Lindenberger |
| 2023 | Tracking Everything Everywhere All at Once | Qianqian Wang |
| 2025 | FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models | Vladimir Kulikov |
Focal Loss in 2017 is the more famous of these in production: the loss function it introduced, originally for the RetinaNet object detector, is now used in dozens of unrelated classification and detection systems.
Beyond the best paper awards, ICCV gives out three other major prizes at each edition.
The Helmholtz Prize (called the Test of Time Award before 2013) recognises ICCV papers from roughly ten years earlier whose impact has held up. It often goes to multiple papers in a single year. Recent winners include "Fast R-CNN" by Ross Girshick and "Delving Deep into Rectifiers" by Kaiming He et al. (both honoured in 2025), "Action Recognition With Improved Trajectories" by Heng Wang and Cordelia Schmid (2023), and "ORB: An efficient alternative to SIFT or SURF" by Rublee, Rabaud, Konolige, and Bradski (2021). The 2017 ceremony recognised an unusually long list, including Sivic and Zisserman's "Video Google," Brown and Lowe's "Recognising Panoramas," and Laptev and Lindeberg's "Space-time Interest Points."
The Azriel Rosenfeld Lifetime Achievement Award honours senior researchers for sustained contributions over a career. Recent recipients include Rama Chellappa (2025), Edward Adelson (2023), Ruzena Bajcsy (2021), Shimon Ullman (2019), Tomaso Poggio (2017), Olivier Faugeras (2015), Jan Koenderink (2013), Thomas Huang (2011), Berthold K. P. Horn (2009), and Takeo Kanade (2007).
The PAMI Distinguished Researcher Award (called the Significant Researcher Award before 2013) is given to two researchers per edition for major mid- or late-career contributions. Pairings have included Michal Irani and David Forsyth (2025), Rama Chellappa and Michael Black (2023), Cordelia Schmid and Pietro Perona (2021), Shree Nayar and William T. Freeman (2019), Luc van Gool and Richard Szeliski (2017), Yann LeCun and David Lowe (2015), Jitendra Malik and Andrew Zisserman (2013), and Richard Hartley and Katsushi Ikeuchi (2011).
The Mark Everingham Prize is given roughly annually (split between ICCV and CVPR) for selfless community service, typically by people who maintained a major dataset, benchmark, or open-source library used by the rest of the field.
ICCV's main program is bookended by one or two days of workshops and tutorials, and recent editions have had between 60 and 100 satellite events. Workshops range from broad theme tracks ("Vision for All Seasons," "Computer Vision for Wildlife Conservation") to dataset and benchmark workshops, to challenge series like SLAM, autonomous driving, or video understanding. Many of these workshops have a competition component with their own leaderboards and prizes; the ICCV 2023 SLAM Challenge, for instance, drew dozens of teams from industry research labs.
Tutorials are aimed at researchers entering a subfield: a typical edition might have a half-day tutorial on neural radiance fields, another on diffusion models for vision, a third on geometric deep learning, and so on. Slides and recordings are usually posted on the conference's CVF page within a few weeks.
There is also an industrial expo running in parallel with the technical program, with booths from companies like Meta, Google, Apple, Amazon, Microsoft, Tesla, NVIDIA, Adobe, and an increasingly long tail of vision-focused startups. The expo doubles as the main recruiting event for a meaningful chunk of the field.
ICCV's influence on the field is hard to overstate, mostly because for many years it was one of the only places where the central conversations of computer vision actually happened. The 1990s ICCV proceedings are essentially the canonical record of multi-view geometry and structure-from-motion as those subjects were being developed; if you read Hartley and Zisserman's textbook, almost every chapter cites at least one ICCV paper. The 2003 prize papers on image parsing, image-based rendering, and pedestrian detection mark the transition into the boosting and probabilistic-graphical-model era. After 2012 the proceedings became the obvious place to track the spread of deep learning into vision: Mask R-CNN at ICCV 2017, SinGAN at ICCV 2019, and Swin Transformer at ICCV 2021 are three of the most cited vision papers of the entire decade.
The Marr Prize itself functions as a useful informal marker. A paper winning the Marr Prize is, in practice, almost a guarantee that it will become a standard reference in its subfield within a year or two, and several Marr Prize winners (Mask R-CNN, Swin Transformer, ControlNet) have crossed over into general-purpose tooling that gets used outside vision research entirely. The Helmholtz Prize, which only gives out awards retrospectively, provides the historical view: looking at the 2013 list, for example, you find Snakes, the bilateral filter, the eight-point algorithm, the Earth Mover's Distance, Tomasi and Kanade's mutual information alignment, and Zhang's flexible camera calibration, basically a catalogue of what every computer vision course taught for a generation.
ICCV, CVPR, and ECCV are the three top peer-reviewed venues in computer vision. They are roughly comparable in selectivity and prestige, and a researcher's CV in vision is usually counted in terms of papers at this trio plus NeurIPS and ICML. There are real differences, though.
| Conference | Frequency | Sponsor | Region | Approximate scale (2023 to 2025) |
|---|---|---|---|---|
| CVPR | Annual, June | IEEE / CVF | Mostly North America | 11,000 to 13,000 submissions, ~25% acceptance |
| ICCV | Biennial, odd years, October to November | IEEE / CVF | Rotates globally | 8,000 to 11,000 submissions, ~25% acceptance |
| ECCV | Biennial, even years, late summer or fall | European CV community | Mostly Europe | 6,000 to 8,000 submissions, ~25% acceptance |
CVPR is typically the largest of the three by submission count, partly because it runs every year and partly because it usually overlaps with the US academic calendar in a way that is convenient for North American groups. ECCV alternates years with ICCV, which is a deliberate scheduling choice: when ECCV is held in even years, ICCV is held in odd years, so the field has roughly two top-tier venues per year (CVPR plus one of the two). Older editions of ECCV are still published as Springer LNCS volumes, while ICCV and CVPR proceedings appear in the IEEE/CVF Open Access library.
In day-to-day practice, none of the three is meaningfully "better" than the others. The same paper would have a roughly similar chance of acceptance at any of them in a given year. The choice of venue is usually driven by deadline timing, location convenience, and which specific area chairs are likely to handle a submission.