Researchers from the University of California Center for Research in Computer Vision (CRCV) had a strong presentation at one of the world’s top computer vision conferences.
The annual Conference on Computer Vision and Patterns Recognition (CVPR), recently held in New Orleans, has been ranked as the fourth best place to publish among all sciences.
The event was organized by Mubarak Shah, Director of CRCV at UCF and Chair Professor of the Board of Trustees, as one of the Conference Chairs.
Shah, along with three other CRCV faculty, Associate Professors of the Department of Computer Science, Chen Chen and Yogesh Rawat, Department of Electrical and Computer Engineering Professor Nazanin Rahnavard, and 32 PhD students and alumni, traveled more than 600 miles to experience the conference in person.
Matthias Mendeta, one of Chen’s doctoral students, along with doctoral student Taojianan Yang and researchers from Tulane University and the University of North Carolina at Charlotte, were finalists for the best paper award at the conference.
Their paper ranked 33research and development Out of 8,161 posts from all over the world.
The paper, “Local Learning Issues: Rethinking Data Heterogeneity in Unified Learning,” focuses on providing insight into improving the effectiveness of Unified Learning, which uses collaborative machine learning while keeping new data local and private for a single device. The goal is to facilitate the development of powerful machine learning models that can be trained without having to access private data.
Mendita says this has influential applications in different fields, such as medicine, where data often cannot be shared with other entities.
“Attending the oral presentations and poster sessions was insightful and inspirational,” Menedita says. “I was grateful to have the opportunity to present our work as an oral presentation during the conference.”
Computer Science PhD student Akash Kumar attended and presented his poster at the conference. His work, Comprehensive Semi-Supervised Learning for Video Motion Detection, explores how researchers can approach a problem using fewer labels.
“Finding out the action of the video is a challenging task because it requires a lot of descriptive and explanatory data, and it is very expensive,” Kumar says. “My approach examines how we can achieve the same level of performance but with much less labeled data, particularly how we can use unlabeled data more efficiently.”
In addition to undergoing high quality research and interacting with leading researchers, the students were able to speak with many of the companies present as well. This was especially important for Kumar, who plans to join the industry after his graduation.
“I met many professors who are experts in their research field, interacted with a lot of companies, and learned about the research industries they are currently focusing on and how they approach these issues in the real world,” says Kumar. “It was nice talking with my senior Ph.D. Students who can guide you on how to do research and make continued progress, because they are in the same boat as us. They helped me look at my research problem from a different perspective.”
Along with Shah, who served as general co-chair of the research conference, Associate Professors Chen and Rawat organized one-day interactive workshops.
Chen’s workshop, “Dynamic Neural Networks Meet Computer Vision”, brought together emerging research in the areas of dynamic optimization of deep neural networks, predictive control, dynamic reasoning for neural codes and computer vision in order to discuss challenges and opportunities open in the future.
Chen was also the lead organizer of the “First International Workshop on FedVision Learning for Computer Vision (FedVision),” a workshop on federal learning and how it can help keep information private.
Rawat organized a workshop entitled “Durability in Sequential Data”. Durability is an important step towards developing reliable systems that can be deployed in the real world. This workshop encouraged researchers to explore the robustness of models against real-world distribution shifts while working on video and language-based sequential data.
He was also part of the organization of the Small Action Challenge, a task focused in a series of challenges aimed at recognizing small or small actions in low-resolution videos that are not clearly visible.
More than 5,500 people attended the conference in person, with nearly 2,000 others joining. It has been on a hiatus for three years thanks to the COVID-19 pandemic.
“We weren’t sure whether or not we should do it in person, but we finally decided to go ahead with an in-person conference,” Shah says. “It was a huge success.”