notesum.ai

Published at November 29

Actions and Objects Pathways for Domain Adaptation in Video Question Answering

cs.CV

cs.CL

Released Date: November 29, 2024

Authors: Safaa Abdullahi Moallim Mohamud¹, Ho-Young Jung²

Aff.: ¹Center for ICT and Automotive Convergence, Kyungpook National University, Daegu, 41566, Republic of Korea; ²Graduate School of Artificial Intelligence, Kyungpook National University, Daegu, 41566, Republic of Korea

Arxiv: http://arxiv.org/pdf/2411.19434v1

	$c\rightarrow m$	$m\rightarrow s$	$s\rightarrow c$		$c\rightarrow m$	$m\rightarrow s$	$s\rightarrow c$
No. Labels: 2	63.64	63.37	66.75	Feature Size: 128	63.42	63.44	66.99
No. Labels: 15	63.67	63.37	66.81	Feature Size: 256	63.67	63.37	66.81
No. Labels: 30	63.45	63.08	66.72	Feature Size: 384	63.23	63.08	66.60
Actions Pathway	62.96	63.31	66.66	Audio & Pathways	62.06	62.81	66.48
Objects Pathway	63.14	63.41	66.75	Text & Pathways	63.67	63.37	66.81
Actions & Objects	63.67	63.37	66.81	A & T & P	62.83	62.98	66.99
Attention	63.67	63.37	66.81	Text Output	61.38	61.72	65.24
No-Attention	62.55	62.35	66.14	Pathways Output	62.89	62.74	65.90