Lightweight TCNs Reduce Emotion Recognition Latency by 12.5x

Temporal Convolutional Networks (TCNs) achieve performance within 1.58 accuracy points of graph-based models while using 79% fewer parameters. This approach reduces classifier latency by approximately 12.5x, making real-time body-based emotion recognition more feasible.