Counting Varying Density Crowds Through Density Guided Adaptive Selection CNN and Transformer Estimation

Chen, Yuehai; Yang, Jing; Chen, Badong; Du, Shaoyi

doi:10.1109/tcsvt.2022.3208714

Cited by 27 publications

(3 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tian et al [17] proposed a pyramid feature aggregation network, which fuses global context features extracted by Transformer with local features to generate high-quality density maps. Song et al [18] designed an aggregated counting network that combines a point-based localization method to improve the network's sensitivity to crowd location; Chen et al [19] designed a new region selection mechanism to classify image regions as high/low density based on combining the output features of the Transformer and the convolutional layers to improve the counting accuracy; Wang et al [20] proposed a spatial context learning network that Convolutional layers with different convolutional kernel sizes are utilized to extract features at the corresponding scales in order to obtain the spatial information required for counting.…”

Section: Counting Methods Based On Density Mapsmentioning

confidence: 99%

CSFNet: A novel crowd counting network for occlusion and scale variation

Xiong,

Li,

Huang

et al. 2024

Preprint

View full text Add to dashboard Cite

The goal of crowd-counting techniques is to estimate the number of people in an image or video in real-time and accurately. In recent years, with the development of deep learning, the accuracy of the crowd-counting task has been improving. However, this task still faces great challenges in crowded scenarios with large individual size variations. To cope with this situation, this paper proposes a new type of crowd-counting network: Context-Scaled Fusion Network. The details include (1) the design of the Multi-Scale Receptive Field Fusion Module (MRFF Module), which employs multiple dilated convolutional layers with different dilatation rates and uses a fusion mechanism to obtain multi-scale hybrid information to generate higher quality feature maps; (2) The Contextual Space Attention Module ( CSA Module) is proposed, which can obtain pixel-level contextual information and combine it with the attention map to enable the model to autonomously learn and pay attention to the important regions to achieve the effect of reducing the counting error. In this paper, we train and test several publicly available and challenging datasets to evaluate the performance of CSFNet. The experimental results show that CSFNet outperforms many SOTA methods on these datasets, demonstrating its superior counting ability and robustness.

show abstract

Section: Counting Methods Based On Density Mapsmentioning

confidence: 99%

CSFNet: A novel crowd counting network for occlusion and scale variation

Xiong,

Li,

Huang

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…SANet (Cao et al 2018) introduced the SSIM ( Wang et al 2004) loss function, which optimizes the similarity between the predicted density map and the ground truth density map based on three indicators: brightness, contrast, and structure. CTASNet (Chen et al 2022) used optimal transport loss and total variation loss to reduce the impact of annotation noise. In this work, we also construct a comprehensive loss function to enhance the accuracy and robustness of the model.…”

Section: Crowd Countingmentioning

confidence: 99%

“…These advantages partially compensate for the limitations of CNNs, enabling the Transformer to better comprehend the relationships between semantic elements in different regions of the entire image. However, Chen et al (2022) found that a pure Transformer is less reliable in sparse crowd regions, and it is difficult to accurately localize and count the target crowds. This is primarily attributed to the self-attention mechanism of Transformers, which excels at capturing global relationships but lacks direct modeling of local receptive fields.…”

Section: Introductionmentioning

confidence: 99%

Global Vision, Local Focus: The Semantic Enhancement Transformer Network for Crowd Counting

Wang,

Zhou,

Chen

2024

Preprint

View full text Add to dashboard Cite

Automatic crowd counting has made significant progress in recent years. However, due to the challenge of multi-scale variations, convolutional neural networks (CNNs) with fixed-size kernels cannot effectively address this difficulty, leading to a severe limitation on counting performance. To alleviate this issue, we propose a semantic enhancement Transformer crowd counting network (named SET) to improve the semantic encoding relationships in crowd scenes. The SET integrates global attention from Transformer, learnable local attention, and inductive bias from CNNs into a counting model. Firstly, we introduce an efficient Transformer encoder to extract low-level global features of crowd scenes. Secondly, we propose a learnable ViTBlock to dynamically learn appropriate weights for different regions, aiding in enhancing the model’s global visual understanding. Finally, to guide the model to focus better on crowd regions, we jointly employ a segmentation attention module and a feature aggregation module to aggregate semantic and spatial features at multiple levels, thus obtaining finer-grained features. We conduct extensive experiments on four challenging datasets, including ShanghaiTech Part A/B, UCF-QNRF, and JHU-CROWD++, achieving excellent results.

show abstract