Last reviewed
May 2, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 4,778 words
Add missing citations, update stale details, or suggest a clearer explanation.
CVPR, the IEEE/CVF Conference on Computer Vision and Pattern Recognition, is the leading annual academic conference for computer vision research. First held in 1983 in Washington, D.C. and run continuously since 1985, it is jointly sponsored by the IEEE Computer Society and, since 2012, the Computer Vision Foundation (CVF). Most of the foundational papers of the deep learning era of computer vision were presented at CVPR or its sister conferences ICCV and ECCV, including ResNet (2016 best paper), DenseNet (2017 best paper), Faster R-CNN, YOLO, and the family of vision transformer models that followed [1][2][3].
By size CVPR is now one of the largest scientific gatherings in artificial intelligence. The 2024 edition in Seattle drew more than 12,000 in-person attendees from 76 countries and reviewed 11,532 paper submissions, of which 2,719 were accepted (an acceptance rate of 23.6%) [4][5]. The 2025 edition in Nashville received 13,008 valid submissions and accepted 2,878 (22.1%) [6]. Industry presence at the conference has grown to match: every major AI lab and most major hardware vendors sponsor the meeting, recruit at it, and present at the affiliated workshops.
| Full name | IEEE/CVF Conference on Computer Vision and Pattern Recognition |
| Abbreviation | CVPR |
| Discipline | Computer vision, pattern recognition, machine learning |
| First held | 1983, Washington, D.C. |
| Founders | Takeo Kanade, Dana H. Ballard |
| Frequency | Annual (since 1985) |
| Sponsors | IEEE Computer Society; Computer Vision Foundation (since 2012) |
| Open access | CVF Open Access (since 2013) |
| 2024 attendance | ~12,000 in person |
| 2024 acceptance rate | 23.6% (2,719 of 11,532) |
| 2025 acceptance rate | 22.1% (2,878 of 13,008) |
| Website | cvpr.thecvf.com |
CVPR began as a one-off symposium organized by Takeo Kanade and Dana H. Ballard, held in Washington, D.C. in June 1983 under the title IEEE Computer Society Conference on Computer Vision and Pattern Recognition [1][7]. The 1983 proceedings, published by IEEE Computer Society Press, contain about a hundred papers and are now mostly of historical interest, dominated by topics such as edge detection, stereo, structure from motion, and early model-based recognition. The conference was not held in 1984, but starting in 1985 it has run every year, almost always in June in North America [1].
For its first quarter century CVPR was a relatively small affair. Through the late 1990s it was common for the conference to receive fewer than a thousand submissions and to accept somewhere between 200 and 400 papers across a single track of orals and posters. Even in 2010, by which time the modern boom in machine learning was well under way, CVPR received 1,724 papers and accepted 461 [8]. The explosion came with the deep learning era: by 2018 submissions had nearly doubled again to 3,359, and by 2024 they had crossed 11,000.
From 1985 to 2010 the conference was sponsored solely by the IEEE Computer Society. In 2011 the University of Colorado Colorado Springs joined as a co-sponsor, and from 2012 onward it has been co-sponsored by the IEEE Computer Society and the Computer Vision Foundation [1]. The conference is administered by a rotating group of volunteers who are elected at the annual meeting of the IEEE Pattern Analysis and Machine Intelligence Technical Committee (PAMI-TC), four years before the conference they will help run.
Takeo Kanade, the founding chair, was at the time at Carnegie Mellon University, where he had moved from Kyoto University in 1980; he later directed the Robotics Institute and remained a regular fixture at the conference for decades. Dana H. Ballard, his co-organizer, was at the University of Rochester and was best known at the time as the co-author with Christopher Brown of the textbook Computer Vision (1982), a book that helped define the boundaries of the field for the first generation of CVPR attendees. Both went on to receive the conference's Azriel Rosenfeld Lifetime Achievement Award (Kanade in 2007, with multiple of his students later receiving similar honors).
The Computer Vision Foundation (CVF) is a non-profit organization registered in Delaware whose stated purpose is "to foster and support research on all aspects of computer vision" [9]. In practice the foundation has two main jobs. First, it co-sponsors the three major CV conferences: CVPR every year, ICCV in odd years, and WACV (the Winter Applications of Computer Vision conference). Second, it operates the CVF Open Access portal, where every paper from a CVF-sponsored conference is hosted as a free PDF in perpetuity, alongside the IEEE Xplore version that sits behind a paywall.
Open access proceedings began with CVPR 2013 and have continued every year since. The arrangement is unusual among IEEE conferences: papers carry the standard IEEE copyright, but CVF is granted permission to host the camera-ready PDFs at no cost to the reader. For a researcher in 2026, this means the entire historical archive of CVPR back to 2013, plus all of ICCV and WACV from the same period, can be searched and downloaded without an institutional subscription. The model has been credited with accelerating the diffusion of computer vision research into adjacent fields and into industry [10].
CVF also occasionally funds travel grants, doctoral consortium awards, and the prize money attached to the conference's various technical awards.
CVPR uses a double-blind peer review process. Authors submit anonymous PDFs through OpenReview (since 2024; CMT was used in earlier years), accompanied by supplementary material, and reviewers are not told the authors' names or affiliations [11]. The page limit for the main text has been eight pages of content plus unlimited references for many years, with a strict line-count requirement enforced by the LaTeX style file.
The review hierarchy goes program chairs at the top, then senior area chairs (SACs), area chairs (ACs), and reviewers. Each paper is assigned three reviewers and at least one AC. The AC writes a meta-review summarizing the reviews and a recommendation; SACs oversee groups of ACs and arbitrate disagreements. Since 2018 the conference has used an "AC triplet" system, in which three ACs jointly review and discuss the borderline papers in a small batch, in an attempt to reduce single-AC variance.
Authors are given a one-page rebuttal window after the initial reviews, which is read by both reviewers and ACs. A 2018 PAMI-TC motion forbids reviewers from requesting substantial new experiments during the rebuttal phase, on the reasoning that the academic resources available in a one-week rebuttal window are too limited to make such requests fair [11].
Program chairs make the final accept/reject decisions and assign accepted papers to one of three presentation categories: oral, highlight (also called spotlight in some years), or poster. Orals are typically capped at around 3% of all submissions, with another 5 to 10% receiving a highlight slot. The remainder are presented as posters. Authors of all accepted papers are required to present a poster regardless of their oral or highlight assignment.
CVPR is famously selective. Acceptance rates have hovered between 22% and 30% for the entire deep learning era, even as raw submissions have grown roughly tenfold. The table below collects published numbers from the conference's official statistics and Paper Copilot's longitudinal tracker [8].
| Year | Submissions | Accepted | Acceptance rate | Orals |
|---|---|---|---|---|
| 2010 | 1,724 | 461 | 26.7% | 78 (4.5%) |
| 2011 | 1,677 | 438 | 26.1% | 59 (3.5%) |
| 2012 | 1,933 | 466 | 24.1% | 48 (2.5%) |
| 2013 | 1,816 | 472 | 26.0% | 60 (3.3%) |
| 2014 | 1,807 | 540 | 29.9% | 104 (5.8%) |
| 2015 | 2,123 | 602 | 28.4% | 71 (3.3%) |
| 2016 | 2,145 | 643 | 30.0% | 83 (3.9%) |
| 2017 | 2,680 | 783 | 29.2% | 71 (2.7%) |
| 2018 | 3,359 | 979 | 29.2% | 70 (2.1%) |
| 2019 | 5,165 | 1,300 | 25.2% | 288 (5.6%) |
| 2020 | 6,424 | 1,467 | 22.8% | 335 (5.2%) |
| 2021 | 7,500 | 1,661 | 22.2% | not reported |
| 2022 | 8,262 | 2,062 | 25.0% | not reported |
| 2023 | 9,155 | 2,360 | 25.8% | not reported |
| 2024 | 11,532 | 2,719 | 23.6% | ~90 oral (3.3%) |
| 2025 | 13,008 | 2,878 | 22.1% | not reported |
The 2024 edition recruited a reviewer pool of around 9,000 people; the 2025 edition expanded that pool to 12,593 reviewers in order to keep the load per reviewer to roughly four papers [6].
The conference's awards program has expanded over the years. Six categories are now standard.
The Best Paper Award is the conference's highest honor for newly published work, given to one or two papers per year. It is selected by an awards committee chaired by the program chairs from a shortlist of nominations. Several past winners have gone on to define entire research areas: the 2009 dark channel haze removal paper from Kaiming He, Jian Sun, and Xiaoou Tang; the 2011 Kinect pose estimation paper from Microsoft Research; the 2016 ResNet paper from He, Zhang, Ren, and Sun; and the 2017 DenseNet paper from Huang, Liu, van der Maaten, and Weinberger. The full list is in the table below.
The Best Student Paper Award goes to a paper whose first author is a student at the time of submission. It typically attracts somewhat newer or higher-risk work than the main Best Paper. The 2017 Best Student Paper, on computational imaging using power-grid flicker (Sheinin et al.), is a good example: the technique would not have made the main award shortlist in a typical year but is widely admired in the computational photography community.
The Longuet-Higgins Prize is CVPR's test-of-time award. It is given each year to one or more CVPR papers from exactly ten years earlier that have had a significant lasting impact on the field. The award is named for H. Christopher Longuet-Higgins, a British theoretical chemist and cognitive scientist whose 1981 paper on the eight-point algorithm for structure from motion is one of the foundational papers of geometric computer vision [12].
Winners include some of the most cited papers in computer vision: the 2019 prize for Deng et al.'s ImageNet database paper from CVPR 2009; the 2024 prize for Girshick et al.'s R-CNN paper from CVPR 2014; and the 2025 prizes for the Fully Convolutional Networks (Long, Shelhamer, Darrell, CVPR 2015) and GoogLeNet (Szegedy et al., CVPR 2015) papers.
The Longuet-Higgins Prize is sometimes informally called CVPR's "Helmholtz Prize," but those are different awards. The Helmholtz Prize is the corresponding test-of-time award at ICCV, not CVPR [13].
The PAMI Young Researcher Award, instituted in 2012, is given to one or two researchers per year who are within seven years of receiving their PhD and have already made substantial contributions to the field. The list of recipients reads as a who's who of mid-career computer vision: Ross Girshick and Julien Mairal in 2017; Andreas Geiger and Kaiming He in 2018; Karen Simonyan in 2019; Jon Barron and Deqing Sun in 2020; Georgia Gkioxari and Phillip Isola in 2021; Bharath Hariharan and Olga Russakovsky in 2022; Judy Hoffman and Christoph Feichtenhofer in 2023; Angjoo Kanazawa and Carl Vondrick in 2024; Saining Xie and Hao Su in 2025.
The Distinguished Researcher Award and the Azriel Rosenfeld Lifetime Achievement Award are both presented at CVPR by the PAMI Technical Committee, in alternating years with the corresponding ICCV ceremony. Past Rosenfeld winners include Takeo Kanade (2007), Berthold Horn (2009), Thomas Huang (2011), Jan Koenderink (2013), Olivier Faugeras (2015), Tomaso Poggio (2017), Shimon Ullman (2019), Ruzena Bajcsy (2021), Edward Adelson (2023), and Rama Chellappa (2025). Distinguished Researcher honorees include Yann LeCun and David Lowe (2015), Luc Van Gool and Richard Szeliski (2017), Shree Nayar and William Freeman (2019), Cordelia Schmid and Pietro Perona (2021), Rama Chellappa and Michael Black (2023), and Michal Irani and David Forsyth (2025).
The Thomas S. Huang Memorial Prize was established at CVPR 2020 in memory of Thomas S. Huang of UIUC, who died in April 2020 after several decades of mentoring computer vision researchers around the world. It has been awarded annually since 2021 to honor researchers recognized as exemplars in research, teaching, mentoring, and service to the community.
The table below lists the Best Paper Award and any co-winners back to 1991. Lists from earlier than that exist but are incomplete in the public record.
| Year | Title | Authors |
|---|---|---|
| 1991 | Face Recognition Using Eigenfaces | M. Turk, A. Pentland |
| 1991 | Robust Dynamic Motion Estimation Over Time | M. J. Black, P. Anandan |
| 1991 | Determining 3D Object Pose Using the Complex Extended Gaussian Image | S. B. Kang, K. Ikeuchi |
| 1994 | Illumination Planning for Object Recognition in Structured Environments | H. Murase, S. Nayar |
| 1996 | What is the Set of Images of an Object Under All Possible Lighting Conditions? | P. Belhumeur, D. Kriegman |
| 1997 | What is a Light Source? | M. Langer, S. Zucker |
| 1997 | Learning Bilinear Models for Two-factor Problems in Vision | W. T. Freeman, J. B. Tenenbaum |
| 2000 | Real-Time Tracking of Non-Rigid Objects using Mean Shift | D. Comaniciu, V. Ramesh, P. Meer |
| 2003 | Object Class Recognition by Unsupervised Scale-Invariant Learning | R. Fergus, P. Perona, A. Zisserman |
| 2005 | Real-Time Non-Rigid Surface Detection | J. Pilet, V. Lepetit, P. Fua |
| 2006 | Putting Objects in Perspective | D. Hoiem, A. Efros, M. Hebert |
| 2007 | Dynamic 3D Scene Analysis from a Moving Vehicle | B. Leibe, N. Cornelis, K. Cornelis, L. Van Gool |
| 2008 | Beyond Sliding Windows: Object Localization by Efficient Subwindow Search | C. Lampert, M. Blaschko, T. Hofmann |
| 2009 | Single Image Haze Removal Using Dark Channel Prior | K. He, J. Sun, X. Tang |
| 2010 | Efficient Computation of Robust Low-Rank Matrix Approximations in the Presence of Missing Data using the L1 Norm | A. Eriksson, A. van den Hengel |
| 2011 | Real-time Human Pose Recognition in Parts from Single Depth Images | J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman, A. Blake |
| 2012 | A Simple Prior-free Method for Non-Rigid Structure-from-Motion Factorization | Y. Dai, H. Li, M. He |
| 2013 | Fast, Accurate Detection of 100,000 Object Classes on a Single Machine | T. Dean, J. Yagnik, M. Ruzon, M. Segal, J. Shlens, S. Vijayanarasimhan |
| 2014 | What Camera Motion Reveals About Shape with Unknown BRDF | M. K. Chandraker |
| 2015 | DynamicFusion: Reconstruction and Tracking of Non-rigid Scenes in Real-Time | R. A. Newcombe, D. Fox, S. M. Seitz |
| 2016 | Deep Residual Learning for Image Recognition (ResNet) | K. He, X. Zhang, S. Ren, J. Sun |
| 2017 | Densely Connected Convolutional Networks (DenseNet) | G. Huang, Z. Liu, L. van der Maaten, K. Q. Weinberger |
| 2017 | Learning from Simulated and Unsupervised Images through Adversarial Training | A. Shrivastava, T. Pfister, O. Tuzel, J. Susskind, W. Wang, R. Webb |
| 2018 | Taskonomy: Disentangling Task Transfer Learning | A. Zamir, A. Sax, W. Shen, L. Guibas, J. Malik, S. Savarese |
| 2019 | A Theory of Fermat Paths for Non-Line-of-Sight Shape Reconstruction | S. Xin, S. Nousias, K. Kutulakos, A. Sankaranarayanan, S. Narasimhan, I. Gkioulekas |
| 2020 | Unsupervised Learning of Probably Symmetric Deformable 3D Objects from Images in the Wild | S. Wu, C. Rupprecht, A. Vedaldi |
| 2021 | GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields | M. Niemeyer, A. Geiger |
| 2022 | Learning to Solve Hard Minimal Problems | P. Hruby, T. Duff, A. Leykin, T. Pajdla |
| 2023 | Visual Programming: Compositional Visual Reasoning Without Training | T. Gupta, A. Kembhavi |
| 2023 | Planning-oriented Autonomous Driving | Y. Hu, J. Yang, L. Chen, K. Li, C. Sima, X. Zhu, S. Chai, S. Du, T. Lin, W. Wang, L. Lu, X. Jia, Q. Liu, J. Dai, Y. Qiao, H. Li |
| 2024 | Generative Image Dynamics | Z. Li, R. Tucker, N. Snavely, A. Holynski |
| 2024 | Rich Human Feedback for Text-to-Image Generation | Y. Liang, J. He, G. Li, P. Li, A. Klimovskiy, N. Carolan, J. Sun, J. Pont-Tuset, S. Young, F. Yang, J. Ke, K. D. Dvijotham, K. M. Collins, Y. Luo, Y. Li, K. J. Kohlhoff, D. Ramachandran, V. Navalpakkam |
| 2025 | VGGT: Visual Geometry Grounded Transformer | J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, D. Novotny |
Sources: CVF awards page [3] and Wikipedia [1].
The Longuet-Higgins Prize, given to a paper from CVPR ten years prior, has gone to many of the most-cited papers in computer vision.
| Year awarded | Original year | Paper | Authors |
|---|---|---|---|
| 2011 | 2001 | Rapid Object Detection using a Boosted Cascade of Simple Features | P. Viola, M. Jones |
| 2013 | 2003 | Object Class Recognition by Unsupervised Scale-Invariant Learning | R. Fergus, P. Perona, A. Zisserman |
| 2015 | 2005 | Histograms of Oriented Gradients for Human Detection | N. Dalal, B. Triggs |
| 2016 | 2006 | Beyond Bags of Features: Spatial Pyramid Matching | S. Lazebnik, C. Schmid, J. Ponce |
| 2018 | 2008 | A Discriminatively Trained, Multiscale, Deformable Part Model | P. Felzenszwalb, D. McAllester, D. Ramanan |
| 2019 | 2009 | ImageNet: A Large-Scale Hierarchical Image Database | J. Deng, W. Dong, R. Socher, L. Li, K. Li, L. Fei-Fei |
| 2020 | 2010 | Secrets of Optical Flow Estimation and Their Principles | D. Sun, S. Roth, M. Black |
| 2021 | 2011 | Real-time Human Pose Recognition in Parts from Single Depth Image | J. Shotton et al. |
| 2022 | 2012 | KITTI Vision Benchmark Suite | A. Geiger, P. Lenz, R. Urtasun |
| 2023 | 2013 | Online Object Tracking: A Benchmark | Y. Wu, J. Lim, M. Yang |
| 2024 | 2014 | Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation (R-CNN) | R. Girshick, J. Donahue, T. Darrell, J. Malik |
| 2025 | 2015 | Fully Convolutional Networks for Semantic Segmentation | J. Long, E. Shelhamer, T. Darrell |
| 2025 | 2015 | Going Deeper with Convolutions (GoogLeNet) | C. Szegedy et al. |
Not every important paper at CVPR has won an award. The conference is also where many of the architectures, techniques, and benchmarks that define modern computer vision were first published. A non-exhaustive list:
As the list shows, CVPR is part of an ecosystem with ICCV, ECCV, NeurIPS, ICML, and ICLR; the boundaries between the venues are not clean and many of the most influential papers cycle between them.
A recent CVPR runs for a full week. The first one or two days are devoted to tutorials and workshops; the next three days are the main conference; and the final day or two are again workshops, plus the doctoral consortium and various meet-the-authors events.
The main conference itself uses a mixed oral/poster format. Roughly two to three percent of accepted papers are given short oral talks, around 10 minutes including Q and A. Another 5 to 10 percent are given highlight or spotlight slots of similar length but in larger sessions. The remainder are presented only as posters. Every accepted paper, including those that were given an oral, must also present a poster, since this is where most discussion actually happens.
Workshops at CVPR have grown into a major sub-conference of their own. The 2024 program included over 130 workshops on topics ranging from autonomous driving and embodied AI to fairness in vision systems and computational pathology. Many workshops run their own paper review and award process and host industry-backed challenges with leaderboards and cash prizes. Tutorials, by contrast, are invited half-day or full-day teaching events; recent topics have included diffusion models, 3D Gaussian splatting, foundation models for video, and the mathematics of equivariant networks.
Industry exhibits and demos run on the trade-show floor in parallel with the main paper sessions. Recent CVPR demo programs have shown live robot demonstrations, real-time generative video, and augmented-reality headsets.
Industry sponsorship is significant. Top-tier sponsors at CVPR 2024 and 2025 included NVIDIA, Google, Meta, Microsoft, Apple, Amazon, Bosch, Toyota Research Institute, Tesla, Bytedance, Alibaba, Baidu, Samsung, Sony, Huawei, and Qualcomm, alongside dozens of smaller startups and academic labs [4][14]. Recruiting at CVPR is intense: a large fraction of attendees are graduate students or recent graduates, and most of the major industrial labs use the conference as their main hiring event of the year.
The sponsorship has come with some complications. The volume of industry-affiliated papers and the prevalence of compute-heavy results has made it harder for purely academic groups to compete on benchmarks, a tension that the program committee has addressed in its recent author guidelines by emphasizing that compute scale alone is not a contribution.
CVPR sits at the top of a small group of computer vision and machine learning conferences that, together, host the field's flagship publications.
| Conference | Frequency | Scope | Approx. acceptance rate (recent) | Open access |
|---|---|---|---|---|
| CVPR | Annual | Computer vision | 22 to 26% | CVF |
| ICCV | Biennial (odd years) | Computer vision | 22 to 27% | CVF |
| ECCV | Biennial (even years) | Computer vision | 25 to 30% | CVF (since 2018) |
| NeurIPS | Annual | All of machine learning | 25 to 27% | OpenReview |
| ICML | Annual | All of machine learning | 25 to 28% | PMLR |
| ICLR | Annual | Representation learning | 30 to 35% | OpenReview |
| WACV | Annual | Applications of computer vision | 30 to 35% | CVF |
| BMVC | Annual | UK and European CV | 30 to 40% | open |
CVPR, ICCV, and ECCV are sometimes grouped as "the big three CV conferences," with the understanding that any author looking to publish a flagship vision paper will target this set. NeurIPS, ICML, and ICLR cover broader machine learning, but their vision tracks are large and compete directly with CVPR for high-profile papers, especially in areas like generative models and self-supervised learning.
A distinctive feature of CVPR among computer science conferences is the size of its in-person community. NeurIPS and ICML are larger by submission count in some recent years, but CVPR consistently draws more on-site attendees: the conference has been the largest in-person AI conference in the world by attendance for several years running [4].
The CVF Open Access portal at openaccess.thecvf.com hosts every paper from every CVF-sponsored conference back to 2013, including CVPR, ICCV, ECCV (from 2018), and WACV. Each paper is available as a free PDF, often alongside the supplementary material and, for recent years, the official video presentation. The same papers are also published in IEEE Xplore, where they sit behind the IEEE paywall but are accessible through most university subscriptions.
For researchers in 2026, the open-access archive is the easiest way to read CVPR papers: the URL pattern is predictable, the PDFs are the camera-ready version, and there is no login required. Many bibliographic management tools, including Google Scholar, link to the CVF version by default.
CVPR rotates between cities, with a strong preference for the United States (the conference's IEEE Computer Society sponsorship is U.S.-based) but occasional visits to Canada and, less frequently, Europe and Asia. Locations are typically chosen four to six years in advance.
| Year | City | Venue |
|---|---|---|
| 2014 | Columbus, Ohio | Greater Columbus Convention Center |
| 2015 | Boston, Massachusetts | Hynes Convention Center |
| 2016 | Las Vegas, Nevada | Caesars Palace |
| 2017 | Honolulu, Hawaii | Hawaii Convention Center |
| 2018 | Salt Lake City, Utah | Salt Palace Convention Center |
| 2019 | Long Beach, California | Long Beach Convention Center |
| 2020 | Originally Seattle (held virtually due to COVID-19) | n/a |
| 2021 | Virtual | n/a |
| 2022 | New Orleans, Louisiana | Ernest N. Morial Convention Center |
| 2023 | Vancouver, British Columbia | Vancouver Convention Centre |
| 2024 | Seattle, Washington | Seattle Convention Center |
| 2025 | Nashville, Tennessee | Music City Center |
| 2026 | Denver, Colorado | Colorado Convention Center (planned) |
Attendance has roughly tracked submissions: the 2020 virtual edition recorded 7,600 registered attendees, the 2024 in-person edition exceeded 12,000, and 2025 in Nashville was on a similar scale [4][15].