VMLH: Efficient Video Moment Location via Hashing

Article Computer Science, Artificial Intelligence

Distilled Siamese Networks for Visual Tracking

Jianbing Shen et al.

Summary: This paper introduces a distilled Siamese tracking framework, which learns small, fast, and accurate trackers through a teacher-student knowledge distillation model from large Siamese trackers. The proposed framework achieves high compression rates and frame rates while maintaining tracking accuracy by utilizing teacher-student distillation and student-student knowledge sharing.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Segmenting Objects From Relational Visual Data

Xiankai Lu et al.

Summary: In this article, the authors propose an attentive graph neural network (AGNN) for pixelwise object segmentation tasks (i.e., automatic video segmentation, image co-segmentation, and few-shot semantic segmentation) in relational visual data. AGNN effectively captures knowledge from the relational visual data through iterative information fusion, leading to accurate object discovery and segmentation.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2022)

Add to Collection

Article Engineering, Electrical & Electronic

Adaptive Region Proposal With Channel Regularization for Robust Object Tracking

Xiankai Lu et al.

Summary: The paper proposes an adaptive region proposal scheme with feature channel regularization for robust object tracking. By integrating correlation filters and adaptively learned region proposals, an enhanced two-stream tracking framework is presented to address tracking failures and scale estimation problems. Extensive experimental validations demonstrate the effectiveness of the proposed method against state-of-the-art tracking algorithms.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Video Moment Localization via Deep Cross-Modal Hashing

Yupeng Hu et al.

Summary: Video moment localization, an important branch of video content analysis, has attracted attention from both industry and academia, facing challenges such as temporal context modeling, intelligent moment candidate generation, and efficiency and scalability. To address these challenges, researchers have proposed a deep end-to-end cross-modal hashing network, demonstrating its superiority through experimental results.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence