Deep-TransVOD: Improved Video Object Detection with Low-Rank Tensor Decomposition
Xinpeng Liu, Yanjie Fang, Weihua Liu · 2024
This study advances Video Object Detection (VOD) through optimization of the TransVOD Lite framework. Our enhancements utilize a Transformer Encoder to improve equivalent mapping and integrate low-rank matrix factorization, thereby eliminating the need for manual post-processing. This upgraded framework, named Deep-TransVOD, is detailed in this paper. It highlights improvements in feature extraction and the implementation of multi-frame input, resulting in a substantial advancement in real-time automated anomaly detection within industrial environments.