Multi-scale 2D Representation Learning for weakly-supervised moment retrieval
This addresses the problem of expensive temporal boundary annotations in video moment retrieval for researchers and practitioners, offering an incremental improvement over existing weakly-supervised methods.
The paper tackles weakly-supervised video moment retrieval by proposing a multi-scale 2D representation learning method to capture temporal dependencies between moment candidates, achieving state-of-the-art results on benchmark datasets Charades-STA and ActivityNet Captions.
Video moment retrieval aims to search the moment most relevant to a given language query. However, most existing methods in this community often require temporal boundary annotations which are expensive and time-consuming to label. Hence weakly supervised methods have been put forward recently by only using coarse video-level label. Despite effectiveness, these methods usually process moment candidates independently, while ignoring a critical issue that the natural temporal dependencies between candidates in different temporal scales. To cope with this issue, we propose a Multi-scale 2D Representation Learning method for weakly supervised video moment retrieval. Specifically, we first construct a two-dimensional map for each temporal scale to capture the temporal dependencies between candidates. Two dimensions in this map indicate the start and end time points of these candidates. Then, we select top-K candidates from each scale-varied map with a learnable convolutional neural network. With a newly designed Moments Evaluation Module, we obtain the alignment scores of the selected candidates. At last, the similarity between captions and language query is served as supervision for further training the candidates' selector. Experiments on two benchmark datasets Charades-STA and ActivityNet Captions demonstrate that our approach achieves superior performance to state-of-the-art results.