Does SpatioTemporal information benefit Two video summarization benchmarks?

TL;DR

It is found that spatio-temporal relationship play a minor role and the question whether these benchmarks adequately model the task of video summarization is raised and whether these benchmarks adequately model the task of video summarization is raised.

Abstract

An important aspect of summarizing videos is understanding the temporal context behind each part of the video to grasp what is and is not important. Video summarization models have in recent years modeled spatio-temporal relationships to represent this information. These models achieved state-of-the-art correlation scores on important benchmark datasets. However, what has not been reviewed is whether spatio-temporal relationships are even required to achieve state-of-the-art results. Previous work in activity recognition has found biases, by prioritizing static cues such as scenes or objects, over motion information. In this paper we inquire if similar spurious relationships might influence the task of video summarization. To do so, we analyse the role that temporal information plays on existing benchmark datasets. We first estimate a baseline with temporally invariant models to see how well such models rank on benchmark datasets (TVSum and SumMe). We then disrupt the temporal order of the videos to investigate the impact it has on existing state-of-the-art models. One of our findings is that the temporally invariant models achieve competitive correlation scores that are close to the human baselines on the TVSum dataset. We also demonstrate that existing models are not affected by temporal perturbations. Furthermore, with certain disruption strategies that shuffle fixed time segments, we can actually improve their correlation scores. With these results, we find that spatio-temporal relationship play a minor role and we raise the question whether these benchmarks adequately model the task of video summarization. Code available at: https://github.com/AashGan/TemporalPerturbSum

Figures (3)

• Figure 1: A visualization, for one video, of the heatmaps of the pairwise differences between the ground truth scores compared with the frame-wise cosine similarity. Here, frames that are visually similar also appear to have low differences in their importance (note, inverse color).
• Figure 2: The Results of the Data Augmentation Experiment. The orange shows the extent to which the models exhibited an improved Correlation Coefficient. As seen here, both PGLSUM and VASNet recorded an improvement, while this improvement was larger in the case of the SumMe dataset.
• Figure 3: Heatmap of Video 32 of the TVSum Dataset