Skip to main content

Table 13 Comparison with state-of-the-art methods (split 1)

From: Gated spatio and temporal convolutional neural network for activity recognition: towards gated multimodal deep learning

Methods UCF-101 HMDB-51
Slow fusion spatiotemporal [8] 36% 36%
Improved dense trajectories (IDT) [20] 85.9% 57.2%
Two stream (averaging fusion) [10] 86.2%
Two stream (SVM fusion) [10] 87.0%
Two stream of good practice [12] 90.2%
Our gating stream + good practice of [12] (VGG-16 gating) 91%
Temporal segment network [23] 93.86% 69.93%
Our gating stream + temporal segment network of [23] (VGG-16 gating) 94.1% 70%