The paper “vCLIMB: A Novel Video Class Incremental Learning Benchmark”, published by Andrés Villa, doctoral student Cenia and internal researcher at KAUST in Saudi Arabia, was received at the Computer Vision and Pattern Recognition 2022 (CVPR) conference, one of the conferences of greater impact on the development and research of artificial intelligence, machine learning, computer vision and deep learning.
The work of Andrés Villa (PhD student in Engineering Sciences with a major in Computer Science at the Pontificia Universidad Católica de Chile) co-authored with Kumail Alhamoud, Juan León Alcázar, Fabian Caba Heilbron, Victor Escorcia and Bernard Ghanem, was accepted in the oral category. This will make it possible to personally expose the process, development and future impacts of his research: “it is a great instance because we can take the work to a platform that can boost visibility astronomically, in addition to finding quality collaborations,” said Villa.
The conference will take place June 19-24 in New Orleans, Louisiana. It will bring together researchers, academics, students and representatives of the world’s leading innovation centers. On this occasion, the event will take place in a hybrid way.
Abstract
Continual learning (CL) is under-explored in the video domain. The few existing works contain splits with imbalanced class distributions over the tasks, or study the problem in unsuitable datasets. We introduce vCLIMB, a novel video continual learning benchmark. vCLIMB is a standardized test-bed to analyze catastrophic forgetting of deep models in video continual learning. In contrast to previous work, we focus on class incremental continual learning with models trained on a sequence of disjoint tasks, and distribute the number of classes uniformly across the tasks. We perform in-depth evaluations of existing CL methods in vCLIMB, and observe two unique challenges in video data. The selection of instances to store in episodic memory is performed at the frame level. Second, untrimmed training data influences the effectiveness of frame sampling strategies. We address these two challenges by proposing a temporal consistency regularization that can be applied on top of memory-based continual learning methods. Our approach significantly improves the baseline, by up to 24% on the untrimmed continual learning task.
