ECO: Efficient Convolution Operators for Tracking

Abstract

In recent years, Discriminative Correlation Filter (DCF) based methods have significantly advanced the state-of-theart in tracking. However, in the pursuit of ever increasing tracking performance, their characteristic speed and realtime capability have gradually faded. Further, the increasingly complex models, with massive number of trainable parameters, have introduced the risk of severe over-fitting. In this work, we tackle the key causes behind the problems of computational complexity and over-fitting, with the aim of simultaneously improving both speed and performance.We revisit the core DCF formulation and introduce: (i) a factorized convolution operator, which drastically reduces the number of parameters in the model; (ii) a compact generative model of the training sample distribution, that significantly reduces memory and time complexity, while providing better diversity of samples; (iii) a conservative model update strategy with improved robustness and reduced complexity. We perform comprehensive experiments on four benchmarks: VOT2016, UAV123, OTB-2015, and Temple-Color. When using expensive deep features, our tracker provides a 20-fold speedup and achieves a 13.0% relative gain in Expected Average Overlap compared to the top ranked method [12] in the VOT2016 challenge. Moreover, our fast variant, using hand-crafted features, operates at 60 Hz on a single CPU, while obtaining 65.0% AUC on OTB-2015. ECOC-COT arXiv:1611.09224v2 [cs.CV]

Online Learning of Discriminative Correlation Filter Bank for Visual Tracking

“…The recent advancement of the performance of the DCF-based tracking algorithm is driven by the reduction of boundary effects [24,25,43] and the adoption of deep features [20,[26][27][28]. When the target moves rapidly and occlusion, the error samples produced by the boundary effect will cause the correlation filter to be weakly discriminated, which results in tracking failure.…”

“…The C-COT model is extremely complex that it sacrifices the real-time capabilities of tracking in exchange for performance standards. Efficient Convolution Operators (ECO) [28] algorithm is the accelerated version of C-COT that optimizes the three aspects of model size, sample set size, and update strategy to achieve acceleration; the tracking speed increases by 20 times compared to that of C-COT. The existing ECO algorithm is the best correlation filter-based tracking algorithm in terms of performance.…”

“…Early DCF-based tracking methods [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29] demonstrate excellent performance in terms of speed. However, correlation filter-based tracking methods with deep features [20,[26][27][28]36,45] have been demonstrated to achieve remarkable performance.…”

“…However, correlation filter-based tracking methods with deep features [20,[26][27][28]36,45] have been demonstrated to achieve remarkable performance. The low-level resolution of the deep feature is high, and the high-level resolution has complete semantic information; the discrimination and invariance of the feature are strong; thus, the tracking performance is evidently improved.…”

“…Conversely, discriminative trackers [5][6][7][8][9] perform tracking by separating the target from the background. In recent years, the existing DCF-based trackers [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29] have demonstrated superior performance in terms of speed on the OTB100 dataset [30]. These trackers are primarily used to learn a DCF for locating the target in a new frame by means of the coordinates of the maximum correlation response.…”