Enhancing Dense Small Object Detection in UAV Images Based on Hybrid Transformer

Transformer-based models have facilitated significant advances in object detection.However,their extensive computational consumption and suboptimal detection of dense small objects curtail their applicability in unmanned aerial vehicle(UAV)imagery.Addressing these limitations,we propose a hybrid transformer-based detector,H-DETR,and enhance it for dense small objects,leading to an accurate and efficient model.Firstly,we introduce a hybrid transformer encoder,which integrates a convolutional neural network-based cross-scale fusion module with the original encoder to handle multi-scale feature sequences more efficiently.Furthermore,we propose two novel strategies to enhance detection performance without incurring additional inference computation.Query filter is designed to cope with the dense clustering inherent in drone-captured images by counteracting similar queries with a training-aware non-maximum suppression.Adversarial denoising learning is a novel enhancement method inspired by adversarial learning,which improves the detection of numerous small targets by counteracting the effects of artificial spatial and semantic noise.Extensive experiments on the VisDrone and UAVDT datasets substantiate the effectiveness of our approach,achieving a significant improvement in accuracy with a reduction in computational complexity.Our method achieves 31.9%and 21.1%AP on the VisDrone and UAVDT datasets,respectively,and has a faster inference speed,making it a competitive model in UAV image object detection.

Enhanced Object Detection and Classification via Multi-Method Fusion

Advances in machine vision systems have revolutionized applications such as autonomous driving,robotic navigation,and augmented reality.Despite substantial progress,challenges persist,including dynamic backgrounds,occlusion,and limited labeled data.To address these challenges,we introduce a comprehensive methodology toenhance image classification and object detection accuracy.The proposed approach involves the integration ofmultiple methods in a complementary way.The process commences with the application of Gaussian filters tomitigate the impact of noise interference.These images are then processed for segmentation using Fuzzy C-Meanssegmentation in parallel with saliency mapping techniques to find the most prominent regions.The Binary RobustIndependent Elementary Features(BRIEF)characteristics are then extracted fromdata derived fromsaliency mapsand segmented images.For precise object separation,Oriented FAST and Rotated BRIEF(ORB)algorithms areemployed.Genetic Algorithms(GAs)are used to optimize Random Forest classifier parameters which lead toimproved performance.Our method stands out due to its comprehensive approach,adeptly addressing challengessuch as changing backdrops,occlusion,and limited labeled data concurrently.A significant enhancement hasbeen achieved by integrating Genetic Algorithms(GAs)to precisely optimize parameters.This minor adjustmentnot only boosts the uniqueness of our system but also amplifies its overall efficacy.The proposed methodologyhas demonstrated notable classification accuracies of 90.9%and 89.0%on the challenging Corel-1k and MSRCdatasets,respectively.Furthermore,detection accuracies of 87.2%and 86.6%have been attained.Although ourmethod performed well in both datasets it may face difficulties in real-world data especially where datasets havehighly complex backgrounds.Despite these limitations,GAintegration for parameter optimization shows a notablestrength in enhancing the overall adaptability and performance of our system.

Confusing Object Detection:A Survey

Confusing object detection(COD),such as glass,mirrors,and camouflaged objects,represents a burgeoning visual detection task centered on pinpointing and distinguishing concealed targets within intricate backgrounds,leveraging deep learning methodologies.Despite garnering increasing attention in computer vision,the focus of most existing works leans toward formulating task-specific solutions rather than delving into in-depth analyses of methodological structures.As of now,there is a notable absence of a comprehensive systematic review that focuses on recently proposed deep learning-based models for these specific tasks.To fill this gap,our study presents a pioneering review that covers both themodels and the publicly available benchmark datasets,while also identifying potential directions for future research in this field.The current dataset primarily focuses on single confusing object detection at the image level,with some studies extending to video-level data.We conduct an in-depth analysis of deep learning architectures,revealing that the current state-of-the-art(SOTA)COD methods demonstrate promising performance in single object detection.We also compile and provide detailed descriptions ofwidely used datasets relevant to these detection tasks.Our endeavor extends to discussing the limitations observed in current methodologies,alongside proposed solutions aimed at enhancing detection accuracy.Additionally,we deliberate on relevant applications and outline future research trajectories,aiming to catalyze advancements in the field of glass,mirror,and camouflaged object detection.

Floating Waste Discovery by Request via Object-Centric Learning

Discovering floating wastes,especially bottles on water,is a crucial research problem in environmental hygiene.Nevertheless,real-world applications often face challenges such as interference from irrelevant objects and the high cost associated with data collection.Consequently,devising algorithms capable of accurately localizing specific objects within a scene in scenarios where annotated data is limited remains a formidable challenge.To solve this problem,this paper proposes an object discovery by request problem setting and a corresponding algorithmic framework.The proposed problem setting aims to identify specified objects in scenes,and the associated algorithmic framework comprises pseudo data generation and object discovery by request network.Pseudo-data generation generates images resembling natural scenes through various data augmentation rules,using a small number of object samples and scene images.The network structure of object discovery by request utilizes the pre-trained Vision Transformer(ViT)model as the backbone,employs object-centric methods to learn the latent representations of foreground objects,and applies patch-level reconstruction constraints to the model.During the validation phase,we use the generated pseudo datasets as training sets and evaluate the performance of our model on the original test sets.Experiments have proved that our method achieves state-of-the-art performance on Unmanned Aerial Vehicles-Bottle Detection(UAV-BD)dataset and self-constructed dataset Bottle,especially in multi-object scenarios.

Rail-Pillar Net:A 3D Detection Network for Railway Foreign Object Based on LiDAR

Aiming at the limitations of the existing railway foreign object detection methods based on two-dimensional(2D)images,such as short detection distance,strong influence of environment and lack of distance information,we propose Rail-PillarNet,a three-dimensional(3D)LIDAR(Light Detection and Ranging)railway foreign object detection method based on the improvement of PointPillars.Firstly,the parallel attention pillar encoder(PAPE)is designed to fully extract the features of the pillars and alleviate the problem of local fine-grained information loss in PointPillars pillars encoder.Secondly,a fine backbone network is designed to improve the feature extraction capability of the network by combining the coding characteristics of LIDAR point cloud feature and residual structure.Finally,the initial weight parameters of the model were optimised by the transfer learning training method to further improve accuracy.The experimental results on the OSDaR23 dataset show that the average accuracy of Rail-PillarNet reaches 58.51%,which is higher than most mainstream models,and the number of parameters is 5.49 M.Compared with PointPillars,the accuracy of each target is improved by 10.94%,3.53%,16.96%and 19.90%,respectively,and the number of parameters only increases by 0.64M,which achieves a balance between the number of parameters and accuracy.

Two-Layer Attention Feature Pyramid Network for Small Object Detection

Effective small object detection is crucial in various applications including urban intelligent transportation and pedestrian detection.However,small objects are difficult to detect accurately because they contain less information.Many current methods,particularly those based on Feature Pyramid Network(FPN),address this challenge by leveraging multi-scale feature fusion.However,existing FPN-based methods often suffer from inadequate feature fusion due to varying resolutions across different layers,leading to suboptimal small object detection.To address this problem,we propose the Two-layerAttention Feature Pyramid Network(TA-FPN),featuring two key modules:the Two-layer Attention Module(TAM)and the Small Object Detail Enhancement Module(SODEM).TAM uses the attention module to make the network more focused on the semantic information of the object and fuse it to the lower layer,so that each layer contains similar semantic information,to alleviate the problem of small object information being submerged due to semantic gaps between different layers.At the same time,SODEM is introduced to strengthen the local features of the object,suppress background noise,enhance the information details of the small object,and fuse the enhanced features to other feature layers to ensure that each layer is rich in small object information,to improve small object detection accuracy.Our extensive experiments on challenging datasets such as Microsoft Common Objects inContext(MSCOCO)and Pattern Analysis Statistical Modelling and Computational Learning,Visual Object Classes(PASCAL VOC)demonstrate the validity of the proposedmethod.Experimental results show a significant improvement in small object detection accuracy compared to state-of-theart detectors.

YOLO-MFD:Remote Sensing Image Object Detection with Multi-Scale Fusion Dynamic Head

Remote sensing imagery,due to its high altitude,presents inherent challenges characterized by multiple scales,limited target areas,and intricate backgrounds.These inherent traits often lead to increased miss and false detection rates when applying object recognition algorithms tailored for remote sensing imagery.Additionally,these complexities contribute to inaccuracies in target localization and hinder precise target categorization.This paper addresses these challenges by proposing a solution:The YOLO-MFD model(YOLO-MFD:Remote Sensing Image Object Detection withMulti-scale Fusion Dynamic Head).Before presenting our method,we delve into the prevalent issues faced in remote sensing imagery analysis.Specifically,we emphasize the struggles of existing object recognition algorithms in comprehensively capturing critical image features amidst varying scales and complex backgrounds.To resolve these issues,we introduce a novel approach.First,we propose the implementation of a lightweight multi-scale module called CEF.This module significantly improves the model’s ability to comprehensively capture important image features by merging multi-scale feature information.It effectively addresses the issues of missed detection and mistaken alarms that are common in remote sensing imagery.Second,an additional layer of small target detection heads is added,and a residual link is established with the higher-level feature extraction module in the backbone section.This allows the model to incorporate shallower information,significantly improving the accuracy of target localization in remotely sensed images.Finally,a dynamic head attentionmechanism is introduced.This allows themodel to exhibit greater flexibility and accuracy in recognizing shapes and targets of different sizes.Consequently,the precision of object detection is significantly improved.The trial results show that the YOLO-MFD model shows improvements of 6.3%,3.5%,and 2.5%over the original YOLOv8 model in Precision,map@0.5 and map@0.5:0.95,separately.These results illustrate the clear advantages of the method.

SMSTracker:A Self-Calibration Multi-Head Self-Attention Transformer for Visual Object Tracking

Visual object tracking plays a crucial role in computer vision.In recent years,researchers have proposed various methods to achieve high-performance object tracking.Among these,methods based on Transformers have become a research hotspot due to their ability to globally model and contextualize information.However,current Transformer-based object tracking methods still face challenges such as low tracking accuracy and the presence of redundant feature information.In this paper,we introduce self-calibration multi-head self-attention Transformer(SMSTracker)as a solution to these challenges.It employs a hybrid tensor decomposition self-organizing multihead self-attention transformermechanism,which not only compresses and accelerates Transformer operations but also significantly reduces redundant data,thereby enhancing the accuracy and efficiency of tracking.Additionally,we introduce a self-calibration attention fusion block to resolve common issues of attention ambiguities and inconsistencies found in traditional trackingmethods,ensuring the stability and reliability of tracking performance across various scenarios.By integrating a hybrid tensor decomposition approach with a self-organizingmulti-head self-attentive transformer mechanism,SMSTracker enhances the efficiency and accuracy of the tracking process.Experimental results show that SMSTracker achieves competitive performance in visual object tracking,promising more robust and efficient tracking systems,demonstrating its potential to providemore robust and efficient tracking solutions in real-world applications.

Learning Discriminatory Information for Object Detection on Urine Sediment Image

In clinical practice,the microscopic examination of urine sediment is considered an important in vitro examination with many broad applications.Measuring the amount of each type of urine sediment allows for screening,diagnosis and evaluation of kidney and urinary tract disease,providing insight into the specific type and severity.However,manual urine sediment examination is labor-intensive,time-consuming,and subjective.Traditional machine learning based object detection methods require hand-crafted features for localization and classification,which have poor generalization capabilities and are difficult to quickly and accurately detect the number of urine sediments.Deep learning based object detection methods have the potential to address the challenges mentioned above,but these methods require access to large urine sediment image datasets.Unfortunately,only a limited number of publicly available urine sediment datasets are currently available.To alleviate the lack of urine sediment datasets in medical image analysis,we propose a new dataset named UriSed2K,which contains 2465 high-quality images annotated with expert guidance.Two main challenges are associated with our dataset:a large number of small objects and the occlusion between these small objects.Our manuscript focuses on applying deep learning object detection methods to the urine sediment dataset and addressing the challenges presented by this dataset.Specifically,our goal is to improve the accuracy and efficiency of the detection algorithm and,in doing so,provide medical professionals with an automatic detector that saves time and effort.We propose an improved lightweight one-stage object detection algorithm called Discriminatory-YOLO.The proposed algorithm comprises a local context attention module and a global background suppression module,which aid the detector in distinguishing urine sediment features in the image.The local context attention module captures context information beyond the object region,while the global background suppression module emphasizes objects in uninformative backgrounds.We comprehensively evaluate our method on the UriSed2K dataset,which includes seven categories of urine sediments,such as erythrocytes(red blood cells),leukocytes(white blood cells),epithelial cells,crystals,mycetes,broken erythrocytes,and broken leukocytes,achieving the best average precision(AP)of 95.3%while taking only 10 ms per image.The source code and dataset are available at https://github.com/binghuiwu98/discriminatoryyolov5.

Robust and Discriminative Feature Learning via Mutual Information Maximization for Object Detection in Aerial Images

Object detection in unmanned aerial vehicle(UAV)aerial images has become increasingly important in military and civil applications.General object detection models are not robust enough against interclass similarity and intraclass variability of small objects,and UAV-specific nuisances such as uncontrolledweather conditions.Unlike previous approaches focusing on high-level semantic information,we report the importance of underlying features to improve detection accuracy and robustness fromthe information-theoretic perspective.Specifically,we propose a robust and discriminative feature learning approach through mutual information maximization(RD-MIM),which can be integrated into numerous object detection methods for aerial images.Firstly,we present the rank sample mining method to reduce underlying feature differences between the natural image domain and the aerial image domain.Then,we design a momentum contrast learning strategy to make object features similar to the same category and dissimilar to different categories.Finally,we construct a transformer-based global attention mechanism to boost object location semantics by leveraging the high interrelation of different receptive fields.We conduct extensive experiments on the VisDrone and Unmanned Aerial Vehicle Benchmark Object Detection and Tracking(UAVDT)datasets to prove the effectiveness of the proposed method.The experimental results show that our approach brings considerable robustness gains to basic detectors and advanced detection methods,achieving relative growth rates of 51.0%and 39.4%in corruption robustness,respectively.Our code is available at https://github.com/cq100/RD-MIM(accessed on 2 August 2024).

A Secure and Cost-Effective Training Framework Atop Serverless Computing for Object Detection in Blasting

The data analysis of blasting sites has always been the research goal of relevant researchers.The rise of mobile blasting robots has aroused many researchers’interest in machine learning methods for target detection in the field of blasting.Serverless Computing can provide a variety of computing services for people without hardware foundations and rich software development experience,which has aroused people’s interest in how to use it in the field ofmachine learning.In this paper,we design a distributedmachine learning training application based on the AWS Lambda platform.Based on data parallelism,the data aggregation and training synchronization in Function as a Service(FaaS)are effectively realized.It also encrypts the data set,effectively reducing the risk of data leakage.We rent a cloud server and a Lambda,and then we conduct experiments to evaluate our applications.Our results indicate the effectiveness,rapidity,and economy of distributed training on FaaS.

Improving Transferable Targeted Adversarial Attack for Object Detection Using RCEN Framework and Logit Loss Optimization

Object detection finds wide application in various sectors,including autonomous driving,industry,and healthcare.Recent studies have highlighted the vulnerability of object detection models built using deep neural networks when confronted with carefully crafted adversarial examples.This not only reveals their shortcomings in defending against malicious attacks but also raises widespread concerns about the security of existing systems.Most existing adversarial attack strategies focus primarily on image classification problems,failing to fully exploit the unique characteristics of object detectionmodels,thus resulting in widespread deficiencies in their transferability.Furthermore,previous research has predominantly concentrated on the transferability issues of non-targeted attacks,whereas enhancing the transferability of targeted adversarial examples presents even greater challenges.Traditional attack techniques typically employ cross-entropy as a loss measure,iteratively adjusting adversarial examples to match target categories.However,their inherent limitations restrict their broad applicability and transferability across different models.To address the aforementioned challenges,this study proposes a novel targeted adversarial attack method aimed at enhancing the transferability of adversarial samples across object detection models.Within the framework of iterative attacks,we devise a new objective function designed to mitigate consistency issues arising from cumulative noise and to enhance the separation between target and non-target categories(logit margin).Secondly,a data augmentation framework incorporating random erasing and color transformations is introduced into targeted adversarial attacks.This enhances the diversity of gradients,preventing overfitting to white-box models.Lastly,perturbations are applied only within the specified object’s bounding box to reduce the perturbation range,enhancing attack stealthiness.Experiments were conducted on the Microsoft Common Objects in Context(MS COCO)dataset using You Only Look Once version 3(YOLOv3),You Only Look Once version 8(YOLOv8),Faster Region-based Convolutional Neural Networks(Faster R-CNN),and RetinaNet.The results demonstrate a significant advantage of the proposed method in black-box settings.Among these,the success rate of RetinaNet transfer attacks reached a maximum of 82.59%.

An object detection approach with residual feature fusion and second-order term attention mechanism

Automatically detecting and locating remote occlusion small objects from the images of complex traffic environments is a valuable and challenging research.Since the boundary box location is not sufficiently accurate and it is difficult to distinguish overlapping and occluded objects,the authors propose a network model with a second-order term attention mechanism and occlusion loss.First,the backbone network is built on CSPDarkNet53.Then a method is designed for the feature extraction network based on an item-wise attention mechanism,which uses the filtered weighted feature vector to replace the original residual fusion and adds a second-order term to reduce the information loss in the process of fusion and accelerate the convergence of the model.Finally,an objected occlusion regression loss function is studied to reduce the problems of missed detections caused by dense objects.Sufficient experimental results demonstrate that the authors’method achieved state-of-the-art performance without reducing the detection speed.The mAP@.5 of the method is 85.8%on the Foggy_cityscapes dataset and the mAP@.5 of the method is 97.8%on the KITTI dataset.

Probability-Enhanced Anchor-Free Detector for Remote-Sensing Object Detection

Anchor-free object-detection methods achieve a significant advancement in field of computer vision,particularly in the realm of real-time inferences.However,in remote sensing object detection,anchor-free methods often lack of capability in separating the foreground and background.This paper proposes an anchor-free method named probability-enhanced anchor-free detector(ProEnDet)for remote sensing object detection.First,a weighted bidirectional feature pyramid is used for feature extraction.Second,we introduce probability enhancement to strengthen the classification of the object’s foreground and background.The detector uses the logarithm likelihood as the final score to improve the classification of the foreground and background of the object.ProEnDet is verified using the DIOR and NWPU-VHR-10 datasets.The experiment achieved mean average precisions of 61.4 and 69.0 on the DIOR dataset and NWPU-VHR-10 dataset,respectively.ProEnDet achieves a speed of 32.4 FPS on the DIOR dataset,which satisfies the real-time requirements for remote-sensing object detection.

CAW-YOLO:Cross-Layer Fusion and Weighted Receptive Field-Based YOLO for Small Object Detection in Remote Sensing

In recent years,there has been extensive research on object detection methods applied to optical remote sensing images utilizing convolutional neural networks.Despite these efforts,the detection of small objects in remote sensing remains a formidable challenge.The deep network structure will bring about the loss of object features,resulting in the loss of object features and the near elimination of some subtle features associated with small objects in deep layers.Additionally,the features of small objects are susceptible to interference from background features contained within the image,leading to a decline in detection accuracy.Moreover,the sensitivity of small objects to the bounding box perturbation further increases the detection difficulty.In this paper,we introduce a novel approach,Cross-Layer Fusion and Weighted Receptive Field-based YOLO(CAW-YOLO),specifically designed for small object detection in remote sensing.To address feature loss in deep layers,we have devised a cross-layer attention fusion module.Background noise is effectively filtered through the incorporation of Bi-Level Routing Attention(BRA).To enhance the model’s capacity to perceive multi-scale objects,particularly small-scale objects,we introduce a weightedmulti-receptive field atrous spatial pyramid poolingmodule.Furthermore,wemitigate the sensitivity arising from bounding box perturbation by incorporating the joint Normalized Wasserstein Distance(NWD)and Efficient Intersection over Union(EIoU)losses.The efficacy of the proposedmodel in detecting small objects in remote sensing has been validated through experiments conducted on three publicly available datasets.The experimental results unequivocally demonstrate the model’s pronounced advantages in small object detection for remote sensing,surpassing the performance of current mainstream models.

MSC-YOLO:Improved YOLOv7 Based on Multi-Scale Spatial Context for Small Object Detection in UAV-View

Accurately identifying small objects in high-resolution aerial images presents a complex and crucial task in thefield of small object detection on unmanned aerial vehicles(UAVs).This task is challenging due to variations inUAV flight altitude,differences in object scales,as well as factors like flight speed and motion blur.To enhancethe detection efficacy of small targets in drone aerial imagery,we propose an enhanced You Only Look Onceversion 7(YOLOv7)algorithm based on multi-scale spatial context.We build the MSC-YOLO model,whichincorporates an additional prediction head,denoted as P2,to improve adaptability for small objects.We replaceconventional downsampling with a Spatial-to-Depth Convolutional Combination(CSPDC)module to mitigatethe loss of intricate feature details related to small objects.Furthermore,we propose a Spatial Context Pyramidwith Multi-Scale Attention(SCPMA)module,which captures spatial and channel-dependent features of smalltargets acrossmultiple scales.This module enhances the perception of spatial contextual features and the utilizationof multiscale feature information.On the Visdrone2023 and UAVDT datasets,MSC-YOLO achieves remarkableresults,outperforming the baseline method YOLOv7 by 3.0%in terms ofmean average precision(mAP).The MSCYOLOalgorithm proposed in this paper has demonstrated satisfactory performance in detecting small targets inUAV aerial photography,providing strong support for practical applications.

Multi-Label Image Classification Based on Object Detection and Dynamic Graph Convolutional Networks

Multi-label image classification is recognized as an important task within the field of computer vision,a discipline that has experienced a significant escalation in research endeavors in recent years.The widespread adoption of convolutional neural networks(CNNs)has catalyzed the remarkable success of architectures such as ResNet-101 within the domain of image classification.However,inmulti-label image classification tasks,it is crucial to consider the correlation between labels.In order to improve the accuracy and performance of multi-label classification and fully combine visual and semantic features,many existing studies use graph convolutional networks(GCN)for modeling.Object detection and multi-label image classification exhibit a degree of conceptual overlap;however,the integration of these two tasks within a unified framework has been relatively underexplored in the existing literature.In this paper,we come up with Object-GCN framework,a model combining object detection network YOLOv5 and graph convolutional network,and we carry out a thorough experimental analysis using a range of well-established public datasets.The designed framework Object-GCN achieves significantly better performance than existing studies in public datasets COCO2014,VOC2007,VOC2012.The final results achieved are 86.9%,96.7%,and 96.3%mean Average Precision(mAP)across the three datasets.

A Simple and Effective Surface Defect Detection Method of Power Line Insulators for Difficult Small Objects

Insulator defect detection plays a vital role in maintaining the secure operation of power systems.To address the issues of the difficulty of detecting small objects and missing objects due to the small scale,variable scale,and fuzzy edge morphology of insulator defects,we construct an insulator dataset with 1600 samples containing flashovers and breakages.Then a simple and effective surface defect detection method of power line insulators for difficult small objects is proposed.Firstly,a high-resolution featuremap is introduced and a small object prediction layer is added so that the model can detect tiny objects.Secondly,a simplified adaptive spatial feature fusion(SASFF)module is introduced to perform cross-scale spatial fusion to improve adaptability to variable multi-scale features.Finally,we propose an enhanced deformable attention mechanism(EDAM)module.By integrating a gating activation function,the model is further inspired to learn a small number of critical sampling points near reference points.And the module can improve the perception of object morphology.The experimental results indicate that concerning the dataset of flashover and breakage defects,this method improves the performance of YOLOv5,YOLOv7,and YOLOv8.In practical application,it can simply and effectively improve the precision of power line insulator defect detection and reduce missing detection for difficult small objects.

Large-field objective lens for multi-wavelength microscopy at mesoscale and submicron resolution

Conventional microscopes designed for submicron resolution in biological research are hindered by a limited field of view,typically around 1 mm.This restriction poses a challenge when attempting to simultaneously analyze various parts of a sample,such as different brain areas.In addition,conventional objective lenses struggle to perform consistently across the required range of wavelengths for brain imaging in vivo.Here we present a novel mesoscopic objective lens with an impressive field of view of 8 mm,a numerical aperture of 0.5,and a working wavelength range from 400 to 1000 nm.We achieved a resolution of 0.74μm in fluorescent beads imaging.The versatility of this lens was further demonstrated through high-quality images of mouse brain and kidney sections in a wide-field imaging system,a confocal laser scanning system,and a two-photon imaging system.This mesoscopic objective lens holds immense promise for advancing multi-wavelength imaging of large fields of view at high resolution.

Masked Autoencoders as Single Object Tracking Learners

Significant advancements have beenwitnessed in visual tracking applications leveragingViT in recent years,mainly due to the formidablemodeling capabilities of Vision Transformer(ViT).However,the strong performance of such trackers heavily relies on ViT models pretrained for long periods,limitingmore flexible model designs for tracking tasks.To address this issue,we propose an efficient unsupervised ViT pretraining method for the tracking task based on masked autoencoders,called TrackMAE.During pretraining,we employ two shared-parameter ViTs,serving as the appearance encoder and motion encoder,respectively.The appearance encoder encodes randomly masked image data,while the motion encoder encodes randomly masked pairs of video frames.Subsequently,an appearance decoder and a motion decoder separately reconstruct the original image data and video frame data at the pixel level.In this way,ViT learns to understand both the appearance of images and the motion between video frames simultaneously.Experimental results demonstrate that ViT-Base and ViT-Large models,pretrained with TrackMAE and combined with a simple tracking head,achieve state-of-the-art(SOTA)performance without additional design.Moreover,compared to the currently popular MAE pretraining methods,TrackMAE consumes only 1/5 of the training time,which will facilitate the customization of diverse models for tracking.For instance,we additionally customize a lightweight ViT-XS,which achieves SOTA efficient tracking performance.