CREST: Convolutional Residual Learning for Visual Tracking

By Yibing Song, Chao Ma, Lijun Gong, Jiawei Zhang, Rynson W. H. Lau, Ming-Hsuan Yang. IEEE International Conference on Computer Vision, 2017

503 citations recorded by the source corpus.

Discriminative correlation filters (DCFs) have been shown to perform superiorly in visual tracking. They only need a small set of training samples from the initial frame to generate an appearance model. However, existing DCFs learn the filters separately from feature extraction, and update these filters using a moving average operation with an empirical weight. These DCF trackers hardly benefit from the end-to-end training. In this paper, we propose the CREST algorithm to reformulate DCFs as a one-layer convolutional neural network. Our method integrates feature extraction, response map generation as well as model update into the neural networks for an end-to-end training. To reduce model degradation during online update, we apply residual learning to take appearance changes into account. Extensive experiments on the benchmark datasets demonstrate that our CREST tracker performs favorably against state-of-the-art trackers.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/c2046fc4744a9d358ea7a8e9c21c92fd58df7a64

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Distractor-aware Siamese Networks for Visual Object Tracking (2018, methodology): “We evaluate the proposed algorithms with numerous fast and state-of-the-art trackers including SiamFC [2], CFNet [33], Staple [1], CSRDCF [19], BACF [11], ECO-HC [3], CREST [29], MDNet [23], CCOT [5], ECO [3], and the baseline tracker SiamRPN [16].”
Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers (2018, methodology): “First, most state-of-the-art trackers spend a signiﬁcant amount of time on the initial training stage [39, 9, 46].”
Learning Aberrance Repressed Correlation Filters for Real-Time UAV Tracking (2019, methodology): “The results are compared with 20 state-of-the-art trackers with both hand-crafted based trackers and deep-based trackers, i.e., KCF [13], DSST [8], SAMF [18], MUSTER [14], BACF [15], SRDCF [7], STAPLE CA [21], MCCT-H [27], STRCF [16], ECO-HC (with gray-scale) [5], ECO [5], CCOT [9], HCF [19], ADNet ”
The Unmanned Aerial Vehicle Benchmark: Object Detection and Tracking (2018, background): “…features (KCF [21], Staple-CA [32], and SRDCFdecon [11]); II) CF trackers with deep features (ECO [9], C-COT [12], HDT [36], CF2 [28], CFNet [43], and PTAV [17]); III) Deep trackers (MDNet [33], SiamFC [5], FCNT [44], SINT [42], MCPF [53], GOTURN [20], ADNet [52], CREST [41], and STCT [45]).”
High Performance Visual Tracking with Siamese Region Proposal Network (2018, methodology): “In this experiment, we compare our method with several representive trackers, including PTAV [11], CREST[31], SRDCF [8], SINT [33], CSR-DCF [23], Siamese-FC [4], Staple [3], CFNet [35] and DSST [9].”
Know Your Surroundings: Exploiting Scene Information for Object Tracking (2020, background): “Another approach is to train or fine-tune a few layers of a deep neural network in the first frame to perform target-background classification [37,8,3,40].”
Target-Aware Deep Tracking (2019, methodology): “the correlation filters based trackers, such as SRDCF [9], Staple [1], MCPF [48], CCOT [10], ECO [7], BACF [17], DRT [38], STRCF [22], and MCCT-H [43]; the non-realtime deep trackers such as MDNet [29], CREST [35], LSART [37], FlowT [52], DSLT [25], MetaSDNet [30], VITAL [36], and DAT [31]; and the ”
Unsupervised Deep Tracking (2019, background): “They incrementally train binary classiﬁcation layers [37, 45, 39] or regression layers [44, 31] based on the initial frame.”
Transformer Meets Tracker: Exploiting Temporal Context for Robust Visual Tracking (2021, methodology): “The recent advances show that the ridge regression can be solved in the deep learning frameworks [42, 32, 7, 3], which avoids the boundary effect in classic DCF trackers.”
Siamese Cascaded Region Proposal Networks for Real-Time Visual Tracking (2018, background): “The comparison with 15 state-of-the-art trackers (SiamRPN [23], DaSiamRPN [59], TRACA [6], ACT [4], BACF [13], ECOHC [7], CREST [42], SiamFC [2], Staple [1], PTAV [11], SINT [45], CFNet [46], SA-Siam [17], HDT [36] and HCFT [32]) is shown in Fig.”
Multi-cue Correlation Filters for Robust Visual Tracking (2018, methodology): “enhanced by using multi-dimensional features [16, 26, 31], part-based strategies [29, 28], adaptive scale [9, 26], longterm framework [32, 17, 45], end-to-end learning [41, 40] and improved filter training methods [20, 30, 33, 14].”
CoTracker: It is Better to Track Together (2023, background): “However, this approximation is crude as points are often strongly correlated ( e.g ., when they belong to the same object).”
Real-Time MDNet (2018, methodology): “Recent techniques based on discriminative correlation filters boost accuracy significantly by incorporating representations from deep neural networks [16, 17, 3, 4, 18].”
Unveiling the Power of Deep Tracking (2018, methodology): “Further, [16] employs DCF as a one-layer CNN for end-to-end training.”
Deep Learning for Visual Tracking: A Comprehensive Survey (2019, methodology): “To achieve the goal of learning generic representations for target modeling and constructing a more robust target models, the main contributions of methods are classified into: i) offline training of CNNs on large-scale datasets for visual tracking [63], [68], [80], [89], [97], [100], [101], [104], ”
Learning Dynamic Memory Networks for Object Tracking (2018, methodology): “Along with the success of convolution neural networks in object recognition and detection, an increasing number of trackers [4, 13, 22, 26, 31] have adopted deep learning models for visual object tracking.”
VITAL: VIsual Tracking via Adversarial Learning (2018, background): “Extensions include, but are not limited to, kernelized correlation filters [23], scale estimation [10], re-detection [37], spatial regularization [12, 14, 9], ADMM optimization [29], sparse representation [32, 42, 33], CNN feature integrations [36, 43, 64, 28] and end-to-end CNN predictions [57, 55,”
AutoTrack: Towards High-Performance Visual Tracking for UAV With Automatic Spatio-Temporal Regularization (2020, methodology): “Compared to deep trackers including ASRCF [31], TADT [51], SiameseFC [15], DSiam [56], MCCT [13], ADNet [16], CFNet [52], DeepSTRCF [14], UDT+ [53], HCF [18], C-COT [17], ECO [20], IBCCF [54], MCPF [19] and CREST [35], AutoTrack with a single CPU exhibits the”