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

Table 4

The impact of the RNNs (LSTM/GRU) on the proposed system’s performance. Training and testing are performed on the same subset (averaged on four subsets).

Method	mAP (%)
	LSTM	GRU
2Stream + WE - VLAD	20.28	20.33
3Stream + WE - VLAD	20.94	20.94
2Stream + WE + VLAD (rnn)	20.74	20.78
3Stream + WE + VLAD (rnn)	21.31	21.35