← Search

Zhangzikang Li

2 accepted papers

2023

Learning Trajectory-Word Alignments for Video-Language Tasks

ICCV 2023poster

In a video, an object usually appears as the trajectory, i.e., it spans over a few spatial but longer temporal patches, that contains abundant spatiotemporal contexts. However, modern Video-Language BERTs (VDL-BERTs) neglect this trajectory characteristic that they usually follow image-language BERT…

Cited by 6PDFScholar
2023

Transforming Visual Scene Graphs to Image Captions

ACL 2023long

We propose to TransForm Scene Graphs into more descriptive Captions (TFSGC). In TFSGC, we apply multi-head attention (MHA) to design the Graph Neural Network (GNN) for embedding scene graphs. After embedding, different graph embeddings contain diverse specific knowledge for generating the words with…