From: Gated spatio and temporal convolutional neural network for activity recognition: towards gated multimodal deep learning
Methods
Accuracy
Spatial streams (three-channel RGB)
36%
Motion streams (three flow fields)
43%
Averaging (model A)
47.5%
Gating network (model C)
48%
Temporal segment network (averaging) [23]
69.93%
Our gating network (model C) + expert network of temporal segment network [23]
70%
