Temporal Aggregate Representations for Long-Range Video Understanding
It addresses future prediction in videos, which is important for applications like robotics and surveillance, but is incremental as it builds on existing techniques like max-pooling and attention.
The paper tackles long-range video understanding by introducing a multi-granular temporal aggregation framework, achieving state-of-the-art results in next action and dense anticipation on datasets like Breakfast, 50Salads, and EPIC-Kitchens.
Future prediction, especially in long-range videos, requires reasoning from current and past observations. In this work, we address questions of temporal extent, scaling, and level of semantic abstraction with a flexible multi-granular temporal aggregation framework. We show that it is possible to achieve state of the art in both next action and dense anticipation with simple techniques such as max-pooling and attention. To demonstrate the anticipation capabilities of our model, we conduct experiments on Breakfast, 50Salads, and EPIC-Kitchens datasets, where we achieve state-of-the-art results. With minimal modifications, our model can also be extended for video segmentation and action recognition.