TAN: Temporal Aggregation Network for Dense Multi-label Action Recognition
This work addresses the challenging problem of predicting multiple action labels per frame in videos for applications like video analysis, representing an incremental improvement over existing methods.
The paper tackles dense multi-label action recognition in videos by proposing the Temporal Aggregation Network (TAN), which decomposes 3D convolutions to reduce model complexity and uses dilated convolutions for multi-scale spatio-temporal aggregation, resulting in outperforming state-of-the-art methods by 5% on Charades and 3% on Multi-THUMOS datasets.
We present Temporal Aggregation Network (TAN) which decomposes 3D convolutions into spatial and temporal aggregation blocks. By stacking spatial and temporal convolutions repeatedly, TAN forms a deep hierarchical representation for capturing spatio-temporal information in videos. Since we do not apply 3D convolutions in each layer but only apply temporal aggregation blocks once after each spatial downsampling layer in the network, we significantly reduce the model complexity. The use of dilated convolutions at different resolutions of the network helps in aggregating multi-scale spatio-temporal information efficiently. Experiments show that our model is well suited for dense multi-label action recognition, which is a challenging sub-topic of action recognition that requires predicting multiple action labels in each frame. We outperform state-of-the-art methods by 5% and 3% on the Charades and Multi-THUMOS dataset respectively.