Introduction
The editorial board of Computational Visual Media is pleased to announce the winners of the 2021 Best Paper Awards. These awards recognize outstanding research contributions published within the journal's 2021 volume, highlighting significant advancements in the field of visual computing.
The Selection Process
The selection process was rigorous and multi-staged. Associate editors initially nominated 11 candidate papers from the 2021 volume. These candidates were then evaluated by a distinguished award committee consisting of:
- Ming C. Lin (University of Maryland, College Park)
- Xin Tong (Microsoft Research Asia)
- Wenping Wang (Texas A&M University)
The committee conducted thorough discussions to assess the original contribution, technical depth, and potential impact of each nominated work.
Awarded Contributions
Best Paper Award: EfficientPose
The top honor was awarded to "EfficientPose: Efficient human pose estimation with neural architecture search" by Wenqiang Zhang, Zhuo Su, Jiaye Lin, and Zhenan Sun. The authors were awarded a US$1,000 prize.
This research addresses the critical challenge of balancing accuracy and computational efficiency in human pose estimation. By employing Neural Architecture Search (NAS), the authors automated the design of the neural network, resulting in a model that achieves state-of-the-art performance while remaining lightweight enough for real-time applications on standard hardware. This reflects a broader industry trend toward optimizing deep learning for real-world performance.
Honorable Mentions
Two additional papers were recognized with Honorable Mentions, each receiving a US$500 prize:
- "Efficient fastest-path computations for road maps" by Renjie Chen and Craig Gotsman. This work focuses on algorithmic scalability, providing faster methods for calculating optimal paths in massive road network datasets.
- "Inferring object properties from human interaction and transferring them to new motions" by Qian Zheng, Boxin Shi, and Gang Pan. This paper explores cross-modal learning, demonstrating how physical properties like weight and friction can be inferred from visual observations of human-object interaction and subsequently applied to generate realistic new animations.
Trends in Visual Computing
The 2021 awards offer a glimpse into the current priorities of the computational visual media community:
- Prioritizing Efficiency: The focus on "efficiency" across multiple winning papers indicates a transition from purely accuracy-driven research to practical, deployable solutions that can run on consumer hardware.
- Automated AI Design: The success of NAS in EfficientPose underscores the growing importance of using AI to design AI itself, moving away from manual feature engineering.
- Physical Realism through Vision: The inference of physical properties from interaction suggests a growing intersection between computer vision and physical simulation, bridging the gap between visual cues and physical properties.
Conclusion
The Best Paper Award Committee congratulates all the winners for their exceptional work. These contributions not only advance the state of the art in computational visual media but also set a high standard for future submissions. The journal continues to welcome high-quality research that pushes the boundaries of efficiency, realism, and automation in visual computing.
Summary updated