Channel Pruning For Visual Tracking

Deep convolutional feature based Correlation Filter trackers have achieved record-breaking accuracy, but the huge computational complexity limits their application. In this paper, we derive the efficient convolution operators (ECO) tracker which obtains the top rank on VOT-2016. Firstly, we introduce a channel pruned VGG16 model to fast extract most representative channels for deep features. Then an Average Feature Energy Ratio method is put forward to select advantageous convolution channels, and an adaptive iterative strategy is designed to optimize object location. Finally, extensive experimental results on four benchmarks OTB-2013, OTB-2015, VOT-2016 and VOT-2017, demonstrate that our tracker performs favorably against the state-of-the-art methods.