. 2023 Jul 13;23(14):6384. doi: 10.3390/s23146384

Table 3.

Architectures applied to HMDB51 using a semi-supervised learning approach.

Variant	Classifier	Learning Rate	Time Step	Method
DINO $_{1}$	3× Convolutions + Medium Grouping + FCN	0.001	1	temporal grouping
DINO $_{2}$	3× Convolutions + FCN	0.1	1	temporal grouping
DINO $_{3}$	1× Convolutions + FCN	0.1	1	temporal grouping
DINO $_{4}$	2× FCN	0.1	1	temporal grouping
DINO $_{5}$	1× FCN	0.1	1	temporal grouping
DINO $_{6}$	1× FCN	0.1	1	temporal grouping
DINO $_{7}$	LSTM + 1× FCN	0.1	4	2D with LSTM
DINO $_{8}$	LSTM + 1× FCN	0.1	1	2D with LSTM