From: Gated spatio and temporal convolutional neural network for activity recognition: towards gated multimodal deep learning
Methods
RGB
Flow
Fusion
Feichtenhofer of late fusion - VGG-M-2048 [22]
74.22%
82.34%
85.94
Feichtenhofer of late fusion - VGG-16 [22]
82.61%
86.25%
90.62
Feature amplification + multiplicative [18]
– %
89.1%
Our gating VGG-16 + expert streams of [12]
79.34%
83.60%
91%