Scaling Human-Object Interaction Recognition in the Video through Zero-Shot Learning

Table 1

Zero-shot HOI recognition mAP on Charades dataset. The model trained with 1A + 2B and tested on 2A + 1B and all data.

Method	mAP (%) on the test set
	All data	Unseen data (2A + 1B)
Chance	1.43	1.45
Compositional [58]	14.32	10.48
SES [68]	13.12	9.56
DEM [75]	11.78	8.97
CC [76]	14.31	10.13
1stream [14]	16.48	11.23
2Stream – SI	17.8	14.83
2Stream + SI	19.5	16.08