RoI Pooling Based Fast Multi-Domain Convolutional Neural Networks for Visual Tracking
Yuanyuan Qin, Shiying He, Yong Hui Zhao, Yuanzhi Gong · 2016
This paper proposes a fast multi-domain convolutional neural networks method (Fast MDNet) for visual tracking.Fast MDNet builds on fast region-based convolutional neural networks (Fast R-CNN) and MDNet to efficiently track arbitrary objects using deep convolutional networks.We introduce a RoI pooling layer which shares full-image convolutional features, thus significantly speed up MDNet.Compared to previous works, Fast MDNet's online tracking rate is 15x faster than MDNet, and it performs favorably against the state-of-the-art methods on large benchmark datasets.