Introduction
This special section of the IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) introduces a curated selection of the highest-rated research from the 2011 Conference on Computer Vision and Pattern Recognition (CVPR). Guest edited by Pedro Felzenszwalb, David Forsyth, Pascal Fua, and Terrance Boult, this collection highlights the "best of the best" from a year that marked a significant turning point in computer vision.
The Selection Process
The CVPR 2011 conference was a massive undertaking, receiving 1,677 full submissions. The editorial team was tasked with selecting the top ~0.5% of these works to be expanded into full journal-quality articles. The rigor of this process is reflected in the conference's statistics: only 59 papers (3.5%) were chosen for oral presentations, with an overall acceptance rate of 26.4%. The papers featured in this special section represent the pinnacle of peer-review ratings and committee awards from that cohort.
Featured Research Themes
The selected works highlight several key technical approaches that defined the research frontier in 2011:
Real-Time Human Pose Estimation
The "Best Paper" award was given to Shotton et al. for their groundbreaking work on real-time human pose recognition from depth images. This research was instrumental in powering the Microsoft Kinect, utilizing Random Forests trained on massive synthetic and captured depth datasets. It demonstrated the potential for robust, real-time interaction on consumer-grade hardware, a revolutionary step at the time.
Large-Scale 3D Reconstruction
Crandall et al. received an honorable mention for their work on Large-Scale Structure from Motion (SfM). By applying discrete-continuous optimization using Markov Random Fields (MRFs), they addressed the complexity of "city-scale" data, combining discrete and continuous mathematical frameworks to improve reconstruction accuracy and scalability.
Semantic Recognition and Contextual Understanding
The "Best Student Paper" award for "Visual Phrases" by Farhadi and Sadeghi signaled a shift toward deeper semantic understanding. Their method captures the relationships between objects in a scene, moving beyond simple object detection to understand the broader context of visual information.
Illumination Invariance
The work of Zhang and Sato, which also received an honorable mention, focused on separating reflective and fluorescent components in images. This research addressed fundamental challenges in illumination invariance, which is critical for robust computer vision in varying lighting conditions.
Historical Context and Enduring Impact
The 2011 committee also awarded the Longuet-Higgins Prize to Paul Viola and Michael Jones for their 2001 paper, "Rapid Object Detection using a Boosted Cascade of Simple Features." By honoring this seminal face detection work a decade later, the committee emphasized that the most valuable research is that which provides enduring tools for the community.
Looking back, the research presented in this special section foreshadowed the major trajectories of the field over the following decade. The focus on large-scale datasets, robust real-time systems, and semantic understanding laid the groundwork for the deep learning revolution that would begin in earnest just a year after this special section was published.
Summary updated