Only Time Can Tell: Discovering Temporal Data for Temporal Modeling
This addresses a benchmarking issue in video understanding for researchers, though it is incremental as it focuses on dataset curation rather than novel modeling.
The paper tackles the problem that many video datasets allow action recognition from single frames, disadvantaging models with strong temporal understanding, by identifying 'temporal classes' where temporal information is necessary through a human annotation experiment, and finds that training on these classes improves generalization to unseen classes.
Understanding temporal information and how the visual world changes over time is a fundamental ability of intelligent systems. In video understanding, temporal information is at the core of many current challenges, including compression, efficient inference, motion estimation or summarization. However, in current video datasets it has been observed that action classes can often be recognized without any temporal information from a single frame of video. As a result, both benchmarking and training in these datasets may give an unintentional advantage to models with strong image understanding capabilities, as opposed to those with strong temporal understanding. In this paper we address this problem head on by identifying action classes where temporal information is actually necessary to recognize them and call these "temporal classes". Selecting temporal classes using a computational method would bias the process. Instead, we propose a methodology based on a simple and effective human annotation experiment. We remove just the temporal information by shuffling frames in time and measure if the action can still be recognized. Classes that cannot be recognized when frames are not in order are included in the temporal Dataset. We observe that this set is statistically different from other static classes, and that performance in it correlates with a network's ability to capture temporal information. Thus we use it as a benchmark on current popular networks, which reveals a series of interesting facts. We also explore the effect of training on the temporal dataset, and observe that this leads to better generalization in unseen classes, demonstrating the need for more temporal data. We hope that the proposed dataset of temporal categories will help guide future research in temporal modeling for better video understanding.