From: Gated spatio and temporal convolutional neural network for activity recognition: towards gated multimodal deep learning
Methods
Accuracy
Spatial streams (three-channel RGB)
72.7%
Motion streams (three flow fields)
76.5%
SVM Fusion (model B)
81.5%
Averaging (model A)
82.7%
Gating network (model C) VGG-16
83%
Gating network (model C) ResNet-50
88.5%