1

Class-incremental learning is one of the most important settings for the study of Continual Learning, as it closely resembles …

Temporal action localization (TAL) is an important task extensively explored and improved for third-person videos in recent years. …

Class-Incremental Learning(CIL) struggles with catastrophic forgetting when learning new knowledge, and Data-Free CIL (DFCIL) is even …

Recent advances in the Active Speaker Detection (ASD) problem build upon a two-stage process – feature extraction and …

The key challenge in neural architecture search (NAS) is designing how to explore wisely in the huge search space. We propose a new NAS …

The recent and increasing interest in video-language research has driven the development of large-scale datasets that enable …

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of daily-life activity video …

Temporal action detection (TAD) is an important yet challenging task in video analysis. Most existing works draw inspiration from image …

Short actions are critical and challenging in the task of action localization. We target this problem and propose a video self-stitching graph network (VSGN), which enhances short action by video self-stitching (VSS) and a cross-scale graph pyramid network (xGPN).

Tackle the problem of network compression and acceleration in a novel perspective: enabling inference on thumbnail images without compromising accuracy. Propose supervised image downscaling, distillation-boosted supervision and feature-mapping regularization.