Probabilistic Regression for Visual Tracking

By Martin Danelljan, Luc Van Gool, Radu Timofte. Computer Vision and Pattern Recognition, 2020

678 citations recorded by the source corpus.

Visual tracking is fundamentally the problem of regressing the state of the target in each video frame. While significant progress has been achieved, trackers are still prone to failures and inaccuracies. It is therefore crucial to represent the uncertainty in the target estimation. Although current prominent paradigms rely on estimating a state-dependent confidence score, this value lacks a clear probabilistic interpretation, complicating its use. In this work, we therefore propose a probabilistic regression formulation and apply it to tracking. Our network predicts the conditional probability density of the target state given an input image. Crucially, our formulation is capable of modeling label noise stemming from inaccurate annotations and ambiguities in the task. The regression network is trained by minimizing the Kullback-Leibler divergence. When applied for tracking, our formulation not only allows a probabilistic representation of the output, but also substantially improves the performance. Our tracker sets a new state-of-the-art on six datasets, achieving 59.8% AUC on LaSOT and 75.8% Success on TrackingNet. The code and models are available at https://github.com/visionml/pytracking.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/6b6d31b022b7984a25fa9ee7fef64086ce7c464d

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

AiATrack: Attention in Attention for Transformer Visual Tracking (2022, neutral): “Tracker SiamRPN++ PrDiMP50 TransT STARK-ST50 KeepTrack AiATrack [34] [12] [8] [58] [41] (Ours) NfS30 [30] 50.”
Unsupervised Person Re-Identification via Multi-Label Classification (2020, background): “(Danelljan et al., 2020) model the label noise and ambiguities through a probabilistic distribution and propose a novel model to estimate an interpretational score for tracking.”
ODTrack: Online Dense Temporal Token Learning for Visual Tracking (2024, methodology): “DCF-based trackers(Danelljan et al. 2019; Bhat et al. 2019; Danelljan, Gool, and Timofte 2020) excel at updating model parameters online using sophisticated optimization techniques, thereby improving the robustness of the tracker.”
Towards More Flexible and Accurate Object Tracking with Natural Language: Algorithms and Benchmark (2021, methodology): “Tracking by Bounding Box The standard trackers begin their tracking procedure based on an initialized BBox in the first frame, including classification based [19,26,44,45], Siamese network based [9, 10, 30, 53, 59], correlation filter based [11, 12, 22, 40], and regression-based [21].”
Is First Person Vision Challenging for Object Tracking? (2020, neutral): “Trackers using an online learning approach [22, 9, 24, 21, 10] generally achieve a below real-time speed.”
ASpanFormer: Detector-Free Image Matching with Adaptive Span Transformer (2022, methodology): “As an elegant framework for uncertainty prediction, some works [15,42,43,44,45] propose to use probabilistic model to jointly explain both flows estimations and their confidence.”
Learning Target Candidate Association to Keep Track of What Not to Track (2021, methodology): “Our method achieves superior results, outperforming LTMU by 6.8% and SuperDiMP by 3.5%.”
Transformer Meets Tracker: Exploiting Temporal Context for Robust Visual Tracking (2021, methodology): “1With the probabilistic IoUNet [9] and a larger search area, our baseline performance is better than the standard DiMP [3].”
Learning Spatio-Temporal Transformer for Visual Tracking (2021, methodology): “6% over PrDiMP [12] respectively, using the same ResNet-50 backbone.”
SeqTrack: Sequence to Sequence Learning for Visual Object Tracking (2023, neutral): “For the classification head network, most trackers adopt stacked convolutional layers with various loss functions, including cross-entropy loss [7, 27], focal loss [51, 55], modified ℓ 2 loss [3, 12], and KL-divergence loss [13].”
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.”
A Review and Comparative Study on Probabilistic Object Detection in Autonomous Driving (2020, background): “…in diﬀerent components of autonomous driving such as semantic segmentation [36]–[38], optical ﬂow [39], [40], depth estimation [35], [41]–[43], visual odometry [44]– [46], image annotation [47]–[50], visual tracking [51], trajectory prediction [52]–[55] and end-to-end perception [56]–[58].”
Deep Learning for Visual Tracking: A Comprehensive Survey (2019, methodology): “The PrDiMP50 [194], DiMP50 [159], and ATOM [149] trackers are the absolute best methods on the LaSOT dataset on all visual attributes.”