Learning the Model Update for Siamese Trackers

By Lichao Zhang, Abel Gonzalez-Garcia, Joost van de Weijer, Martin Danelljan, Fahad Shahbaz Khan. IEEE International Conference on Computer Vision, 2019

385 citations recorded by the source corpus.

Siamese approaches address the visual tracking problem by extracting an appearance template from the current frame, which is used to localize the target in the next frame. In general, this template is linearly combined with the accumulated template from the previous frame, resulting in an exponential decay of information over time. While such an approach to updating has led to improved results, its simplicity limits the potential gain likely to be obtained by learning to update. Therefore, we propose to replace the handcrafted update function with a method which learns to update. We use a convolutional neural network, called UpdateNet, which given the initial template, the accumulated template and the template of the current frame aims to estimate the optimal template for the next frame. The UpdateNet is compact and can easily be integrated into existing Siamese trackers. We demonstrate the generality of the proposed approach by applying it to two Siamese trackers, SiamFC and DaSiamRPN. Extensive experiments on VOT2016, VOT2018, LaSOT, and TrackingNet datasets demonstrate that our UpdateNet effectively predicts the new target template, outperforming the standard linear update. On the large-scale TrackingNet dataset, our UpdateNet improves the results of DaSiamRPN with an absolute gain of 3.9% in terms of success score.

Subfields: Computer Vision and Pattern Recognition, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/157ff6e216985911cc2f9775155d2a424ba2984b

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

In Electrical and Electronic Engineering

Citations in the corpus

Is First Person Vision Challenging for Object Tracking? (2020, positive): “Trackers based on siamese networks [8, 91, 36, 15, 48, 80, 90, 92] emerge as the fastest trackers, thanks to their non-reliance of online adaptation mechanisms.”
Know Your Surroundings: Exploiting Scene Information for Object Tracking (2020, background): “Moreover, a number of attempts have been made to integrate background information into the appearance model in Siamese trackers [52,30,51].”
Deformable Siamese Attention Networks for Visual Object Tracking (2020, neutral): “…0.276 DaSiamRPN [44] 0.61 0.22 0.411 0.59 0.28 0.383 SPM [34] 0.62 0.21 0.434 - - - SiamMask-Opt [36] 0.67 0.23 0.442 0.64 0.30 0.387 UpdateNet [42] 0.61 0.21 0.481 - - 0.393 GFS-DCF [40] - - - 0.51 0.14 0.397 ATOM [6] - - - 0.59 0.20 0.401 SiamRPN++ [24] 0.64 0. forms all previous methods…”
Transformer Meets Tracker: Exploiting Temporal Context for Robust Visual Tracking (2021, methodology): “In other tracking frameworks with update mechanisms [20, 39, 8, 58, 60, 3], previous prediction results are collected to incrementally update the tracking model.”
AutoTrack: Towards High-Performance Visual Tracking for UAV With Automatic Spatio-Temporal Regularization (2020, background): “Currently, the state-of-theart deep trackers [14, 37, 38, 39, 40, 41] are mostly built on siamese neural network [15].”
Deep Learning for Visual Tracking: A Comprehensive Survey (2019, methodology): “20GHz CPU, Titan Xp GPU N/A N/A 43 CM UpdateNet [189] AlexNet LaSOT DAF N/A Python PyTorch N/A CM SPM [153] AlexNet, SiameseRPN, RelationNet ImageNet, ILSVRC-VID, YTBB, ILSVRC-DET, MSCOCO, CityPerson, WiderFace DAF N/A, P100 GPU N/A N/A 120 OS FICFNet [141] AlexNet ImageNet, ILSVRC-VID DAF Intel I7 ”
AiATrack: Attention in Attention for Transformer Visual Tracking (2022, methodology): “A straightforward solution is to update the reference features by generation [63] or ensemble [19,53,58].”
Towards More Flexible and Accurate Object Tracking with Natural Language: Algorithms and Benchmark (2021, background): “Recently, some researchers begin to collect the tracking results which can be used to dynamically update the target template and attain better results [62, 67].”
ODTrack: Online Dense Temporal Token Learning for Visual Tracking (2024, methodology): “Up-dateNet(Zhang et al. 2019) introduces an adaptive updating strategy, which utilizes a custom network to fuse accumulated templates and generate a weighted updated template feature for visual tracking.”
Learning Spatio-Temporal Transformer for Visual Tracking (2021, methodology): “Previous Siamese trackers [25, 51, 14, 6] only exploit the spatial information for tracking, while online methods [54, 57, 9, 3] use historical predictions for model updates.”
Probabilistic Regression for Visual Tracking (2020, methodology): “The aforementioned conﬁdence-based regression strategy is shared by the previously dominant Discriminative Correlation Filter (DCF) paradigm [5, 10, 12, 15, 22, 31, 40] and the more recent Siamese trackers [2, 17, 28, 29, 45, 51].”
Transformer Tracking (2021, methodology): “Thus, previous models have to improve the nonlinear representation ability by introducing fashion structures [21, 44, 48], using additional modules [7, 46, 13], designing effective online updaters [2, 47, 10], to name a few.”
STMTrack: Template-free Visual Tracking with Space-time Memory Networks (2021, methodology): “We test our tracker on the testing set of LaSOT [5], and compare it with twelve competitive methods: LTMU [3], DiMP-50 [1], Occean [21], SiamFC++ [18], GlobalTrack [8], SiamCAR [6], ATOM [4], SiamBAN [2], SiamRPN++ [10], UpdateNet [20], ROAM++ [19], and VITAL [16].”