CVPR 2023: Advanced Computer Vision Research Leads to Record Paper Submissions

When we chose Vancouver for CVPR 2023, we didn’t quite realize how fitting a location it would be. Known as the “Hollywood of the North” due to numerous television shows and movies filmed there, the city projects an impressive, larger-than life-quality—a fitting backdrop for this year’s impressive CVPR program.

In 2023, we had a record number of paper submissions, 9,155 to be exact. To put that into perspective, last year we had a total of 8,162 submissions, so a 12 percent increase in submissions in just one year. Of this astonishing number of submitted papers, the CVPR 2023 Program Committee accepted just 2,359 papers, an acceptance rate of only 25.8 percent, delivering a program of the highest quality, one that sets the bar for what’s next in our industry.

Take, for instance, one of the accepted papers, “A New Benchmark: On the Utility of Synthetic Data with Blender for Bare Supervised Learning and Downstream Domain Adaptation[1].” This research addresses the challenge of large-scale labeled data training, which can be impractical when annotating data for all tasks of all domains of interest, labor-intensive and not completely accurate. Through the use of synthetic data, researchers have found a viable alternative for data training and developed a new industry benchmark for large-scale synthetic-to-real image classification.

Or consider “Implicit Identity Leakage: The Stumbling Block to Improving Deepfake Detection Generalization.” This research explores the unexpected learned identity representation on images by binary classifiers, a phenomenon known as the Implicit Identity Leakage, and designs a method, coined the ID-unaware Deepfake Detection Model, to reduce the occurrence of the phenomenon, which in testing has outperformed other state-of-the-art approaches.

Or review “A Light Weight Model for Active Speaker Detection.” This research explores the challenges in improving active speaker detection, which is crucial in applications like speaker diarization, speaker tracking, and automatic video editing, but to date employed improvement methods that require a high consumption of memory and computational power, making them difficult to be applied in resource-limited scenarios. Leveraging a lightweight active speaker detection architecture with low computational complexity requiring significantly lower resource costs, researchers were able to achieve competitive performance for active speaker detection when matched with more complex models.

These are just three papers that serve as a small sampling of what CVPR 2023 will have to offer; they hint at the wide range of developments happening in the fields of computer vision and pattern recognition. And to think, CVPR 2023 attendees will be able to experience this type of discovery through the 2,356 more papers that will be presented at the event.

It’s clear. CVPR 2023 is going to be big. The number of papers, the quality of the research, and the continual growth in the industry’s desire to present at the conference, demonstrate that this is THE premier event for professionals in the field of computer vision and pattern recognition. So, we invite you to join us in Vancouver, where the stage is set for the field’s preeminent conference. Register today.