CCST: crowd counting with swin transformer

Li, Bo; Zhang, Yong; Xu, Haihui; Yin, Baocai

doi:10.1007/s00371-022-02485-3

Cited by 19 publications

(13 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The traffic scenario of this dataset is highly complex, and the scale of the counting object varies widely. The experimental comparison shows that our method is superior to the other state-of-the-art methods (Li et al, , 2020(Li et al, , 2023Fu et al, 2023).…”

Section: Comparisons With State-of-the-artmentioning

confidence: 87%

“…as the targets of two-channel counting in this dataset. The experimental results show that our method can classify and count cars and busses at the same time, and it is better than the most advanced counting methods (Li et al, , 2020(Li et al, , 2023Fu et al, 2023).…”

Section: Comparisons With State-of-the-artmentioning

confidence: 93%

“…It is a challenging dataset with severe object occlusion and different perspectives. The existing methods, such as Li et al (2018 , 2020 , 2023) and Fu et al (2023) , have performed well. We perform counting experiments on the existing methods by using images of people and vehicles respectively, while our method can simultaneously count people and vehicles in two channels.…”

Section: Methodsmentioning

confidence: 99%

See 2 more Smart Citations

Counting dense object of multiple types based on feature enhancement

Fu,

Min,

Sheng

et al. 2024

Front. Neurorobot.

View full text Add to dashboard Cite

IntroductionAccurately counting the number of dense objects in an image, such as pedestrians or vehicles, is a challenging and practical task. The existing density map regression methods based on CNN are mainly used to count a class of dense objects in a single scene. However, in complex traffic scenes, objects such as vehicles and pedestrians usually exist at the same time, and multiple classes of dense objects need to be counted simultaneously.MethodsTo solve the above issues, we propose a new multiple types of dense object counting method based on feature enhancement, which can enhance the features of dense counting objects in complex traffic scenes to realize the classification and regression counting of dense vehicles and people. The counting model consists of the regression subnet and the classification subnet. The regression subnet is primarily used to generate two-channel predicted density maps, mainly including the initial feature layer and the feature enhancement layer, in which the feature enhancement layer can enhance the classification features and regression counting features of dense objects in complex traffic scenes. The classification subnet mainly supervises classifying dense vehicles and people into two feature channels to assist the regression counting task of the regression subnets.ResultsOur method is compared on VisDrone+ datasets, ApolloScape+ datasets, and UAVDT+ datasets. The experimental results show that the method counts two kinds of dense objects simultaneously and outputs a high-quality two-channel predicted density map. The counting performance is better than the state-of-the-art counting network in dense people and vehicle counting.DiscussionIn future work, we will further improve the feature extraction ability of the model in complex traffic scenes to classify and count a variety of dense objects such as cars, pedestrians, and non-motor vehicles.

show abstract

Section: Comparisons With State-of-the-artmentioning

confidence: 87%

Section: Comparisons With State-of-the-artmentioning

confidence: 93%

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Counting dense object of multiple types based on feature enhancement

Fu,

Min,

Sheng

et al. 2024

Front. Neurorobot.

View full text Add to dashboard Cite

show abstract

“…This is achieved by adaptively fusing multi-level features and exchanging information to prevent an uneven distribution of crowd density. The results showed some improvement in the counting performance of this network [14]. Park K B and other professionals targeted the improvement of segmentation polyp effectiveness by combining EfficientNet and ST to improve the accuracy and robustness of medical segmentation, maintaining global semantics to ensure the scalability and generalization of polyp segmentation.…”

Section: Related Workmentioning

confidence: 88%

3D Target Detection Incorporating Point Cloud Columnarization and Attention Mechanisms in Intelligent Driving Systems

Wang,

Zhang

2024

IEEE Access

View full text Add to dashboard Cite

One crucial problem with intelligent driving systems is 3D target detection. Point cloud data has several uses in the realm of perception and is a vital source of information. To improve the accuracy and robustness of target detection, the study designed a point cloud columnarized network structure based on the PointPillars algorithm. This structure reduces the dimensionality and noise of the data and improves the efficiency of target detection. The Swin Transformer algorithm is utilized to enhance the network structure, enabling the utilization of spatial information from point cloud data for precise detection and localization of 3D targets. The results indicated that the improved model predicted frames and real frames had smaller offset angles and smaller errors, and fit better compared to the PointPillars algorithm. The memory usage of the improved model graphics card was 1253MB, and the running speed was 0.033s, compared with the PointPillars algorithm the memory usage of the graphics card was reduced by 14MB, and the running speed was improved by 0.003s. The first 0.03s of the target detection of the PointPillars model had the most deviation, and the deviation was generally 0.03m. The reason was that PointPillars algorithm is not capable of handling occlusion, small or dense targets well enough to produce errors.The detection error distribution of the improved model was concentrated around 0.01s, and the average deviation was 0.018m, which reduced the deviation by nearly 55.7% compared to the PointPillars model. The enhanced technique enhances the driving system's safety and perception capacity while precisely identifying target items on the road. This is of great significance to promote the development and practical application of intelligent driving systems.

show abstract

“…Taking inspiration from feature selection and the divide-andconquer approach (Chen et al, 2021;Dai et al, 2021;Li et al, 2023), this paper introduces a Feature Adaptive Fusion Regression Head (CRMHead). The core idea of this method is to adaptively select the appropriate feature layers for targets of different scales.…”

Section: Crmheadmentioning

confidence: 99%

FCFormer: fish density estimation and counting in recirculating aquaculture system

Zhu,

Yang,

Yang

et al. 2024

Front. Mar. Sci.

View full text Add to dashboard Cite

In intelligent feeding recirculating aquaculture system, accurately estimating fish population and density is pivotal for management practices and survival rate assessments. However, challenges arise due to mutual occlusion among fish, rapid movement, and complex breeding environments. Traditional object detection methods based on convolutional neural networks (CNN) often fall short in fully addressing the detection demands for fish schools, especially for distant and small targets. In this regard, we introduce a detection framework dubbed FCFormer (Fish Count Transformer). Specifically, the Twins-SVT backbone network is employed first to extract global features of fish schools. To further enhance feature extraction, especially in the fusion of features at different levels, a Bi-FPN aggregation network model with a CAM Count module is incorporated (BiCC). The CAM module aids in focusing more on critical region features, thus rendering feature fusion more cohesive and effective. Furthermore, to precisely predict density maps and elevate the accuracy of fish counting, we devised an adaptive feature fusion regression head: CRMHead. This approach not only optimizes the feature fusion process but also ensures superior counting precision. Experimental results shown that the proposed FCFormer network achieves an accuracy of 97.06%, with a mean absolute error (MAE) of 6.37 and a root mean square error (MSE) of 8.69. Compared to the Twins transformer, there's a 2.02% improvement, outperforming other transformer-based architectures like CCTrans and DM_Count. The presented FCFormer algorithm can be effectively applied to fish density detection in intelligent feeding recirculating aquaculture system, offering valuable input for the development of intelligent breeding management systems.

show abstract

CCST: crowd counting with swin transformer

Cited by 19 publications

References 63 publications

Counting dense object of multiple types based on feature enhancement

Counting dense object of multiple types based on feature enhancement

3D Target Detection Incorporating Point Cloud Columnarization and Attention Mechanisms in Intelligent Driving Systems

FCFormer: fish density estimation and counting in recirculating aquaculture system

Contact Info

Product

Resources

About