Multi-Moments in Time: Learning and Interpreting Models for Multi-Action Video Understanding
This addresses the limitation of single-label datasets in video action recognition, enabling more accurate and comprehensive models for applications in video analysis.
The authors tackled the problem of multi-action video understanding by introducing the Multi-Moments in Time dataset (M-MiT) with over two million labels for one million videos, and they provided baseline results with improved methods for model interpretation and transfer learning.
Videos capture events that typically contain multiple sequential, and simultaneous, actions even in the span of only a few seconds. However, most large-scale datasets built to train models for action recognition in video only provide a single label per video. Consequently, models can be incorrectly penalized for classifying actions that exist in the videos but are not explicitly labeled and do not learn the full spectrum of information present in each video in training. Towards this goal, we present the Multi-Moments in Time dataset (M-MiT) which includes over two million action labels for over one million three second videos. This multi-label dataset introduces novel challenges on how to train and analyze models for multi-action detection. Here, we present baseline results for multi-action recognition using loss functions adapted for long tail multi-label learning, provide improved methods for visualizing and interpreting models trained for multi-label action detection and show the strength of transferring models trained on M-MiT to smaller datasets.