notesum.ai

Published at November 20

DATTA: Domain-Adversarial Test-Time Adaptation for Cross-Domain WiFi-Based Human Activity Recognition

cs.CV

cs.AI

cs.ET

Released Date: November 20, 2024

Authors: Julian Strohmayer¹, Rafael Sterzinger¹, Matthias Wödlinger, Martin Kampel¹

Aff.: ¹Computer Vision Lab, TU Wien

Arxiv: http://arxiv.org/abs/2411.13284v1

Model	A	R	ACC $\uparrow$	F1-Score $\uparrow$
$W$ (baseline)	-		38.69 $\pm$ 3.17	40.62 $\pm$ 2.93
$W$ (baseline)	✓		47.75 $\pm$ 4.31	49.32 $\pm$ 4.45
$W_{\text{TTA}}$	-	-	42.76 $\pm$ 4.21	44.89 $\pm$ 3.93
$W_{\text{TTA}}$	✓	-	51.70 $\pm$ 4.40	53.20 $\pm$ 4.26
$W_{\text{DAT}}$	-		39.54 $\pm$ 2.80	42.02 $\pm$ 2.62
$W_{\text{DAT}}$	✓		64.53 $\pm$ 1.87	65.66 $\pm$ 1.85
$W_{\text{DATTA}}$	-	-	45.26 $\pm$ 5.83	48.23 $\pm$ 5.11
	✓	-	65.90 $\pm$ 1.53	67.29 $\pm$ 1.43
	✓	✓	66.92 $\pm$ 1.54	68.13 $\pm$ 1.45