Uniformity in Heterogeneity: Diving Deep into Count Interval Partition for Crowd Counting

Wang, Chang‐An; Song, Qingyu; Zhang, Boshen; Wang, Yabiao; Tai, Ying; Hu, Xuyi; Wang, Chengjie; Li, Jilin; Ma, Jiayi; Wu, Yang

doi:10.1109/iccv48922.2021.00322

Cited by 28 publications

(20 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead of tackling the counting task through different learning frameworks or strategies, recent methods [66], [67], [68], [69], [70], [71], [72], [73], [74], [75] payed attention at the way of supervisions. For example, Sravya et al proposed a bin loss [68] to enable the data-distribution aware optimization, which helped to address the domain variation challenges from different crowd data source.…”

Section: Learn To Count With Different Supervisionsmentioning

confidence: 99%

“…Recently, [73] proposed distribution matching loss to tackle the weakened generalizability of the Gaussian smoothed density map. Moreover, Wang et al [74] treated the counting with density map as a classification problem, where a Cross-Entropy loss was used to classify each patch into certain intervals.…”

Section: Learn To Count With Different Supervisionsmentioning

confidence: 99%

“…Our proposed L DCD also helps to reduce the original MCNN [18] by 6.4%, the CSRNet [33] by 2.7%, and the CACC [19] by 2.3% over average MAE on two datasets. Please note that, we did not compare with other loss functions that were proposed in recent crowd counting model [69], [71], [70], [73], [74], [72]. Because those methods are not pure density map regression based methods, it is unfair to compare.…”

Section: Ablation Studymentioning

confidence: 99%

See 2 more Smart Citations

Counting with Adaptive Auxiliary Learning

Meng¹,

Bridge²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

This paper proposes an adaptive auxiliary task learning based approach for object counting problems. Unlike existing auxiliary task learning based methods, we develop an attention-enhanced adaptively shared backbone network to enable both task-shared and task-tailored features learning in an end-to-end manner. The network seamlessly combines standard Convolution Neural Network (CNN) and Graph Convolution Network (GCN) for feature extraction and feature reasoning among different domains of tasks. Our approach gains enriched contextual information by iteratively and hierarchically fusing the features across different task branches of the adaptive CNN backbone. The whole framework pays special attention to the objects' spatial locations and varied density levels, informed by object (or crowd) segmentation and density level segmentation auxiliary tasks. In particular, thanks to the proposed dilated contrastive density loss function, our network benefits from individual and regional context supervision in terms of pixel-independent and pixel-dependent feature learning mechanisms, along with strengthened robustness. Experiments on seven challenging multidomain datasets demonstrate that our method achieves superior performance to the state-of-the-art auxiliary task learning based counting methods. Our code is made publicly available at: https: //github.com/smallmax00/Counting With Adaptive Auxiliary

show abstract

Section: Learn To Count With Different Supervisionsmentioning

confidence: 99%

Section: Learn To Count With Different Supervisionsmentioning

confidence: 99%

Section: Ablation Studymentioning

confidence: 99%

See 1 more Smart Citation

Counting with Adaptive Auxiliary Learning

Meng¹,

Bridge²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Crowd counting has recently been one of the popular tasks in computer vision. Recent developed methods [1,2,3] and datasets [4,5,6] have been introduced to tackle the counting task with thousands of targets. However, in real-world scenarios, these supervised methods usually learn to count through a training process that requires an extensive annotation of densely populated points in thousands of images.…”

Section: Introductionmentioning

confidence: 99%

Self-supervised Domain Adaptation in Crowd Counting

Nguyen¹,

Truong²,

Huang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Self-training crowd counting has not been attentively explored though it is one of the important challenges in computer vision. In practice, the fully supervised methods usually require an intensive resource of manual annotation. In order to address this challenge, this work introduces a new approach to utilize existing datasets with ground truth to produce more robust predictions on unlabeled datasets, named domain adaptation, in crowd counting. While the network is trained with labeled data, samples without labels from the target domain are also added to the training process. In this process, the entropy map is computed and minimized in addition to the adversarial training process designed in parallel. Experiments on Shanghaitech, UCF_CC_50, and UCF-QNRF datasets prove a more generalized improvement of our method over the other state-of-the-arts in the cross-domain setting.

show abstract

“…2. Our choice of using center maps for representing humans under occlusion is inspired by crowd-counting literature [44,56,62] and recent works in detection [8,71,72]. Our method, OCHMR, predicts the output mesh from the input image for the person of interest in the subject-specific local center-map.…”

Section: Introductionmentioning

confidence: 99%

Occluded Human Mesh Recovery

Khirodkar¹,

Tripathi²

2022

Preprint

View full text Add to dashboard Cite

ImageCenter Map (Ours) OCHMR (Ours) ROMP Figure 1. To handle severe person-person occlusion, our proposed method OCHMR conditions the deep network on image spatial context using predicted body centermaps. OCHMR is trained using multi-person mesh interpenetration and depth ordering losses. In comparison to bottom-up ROMP [58], top-down OCHMR outputs pixel aligned mesh estimates for each individual under occlusion.

show abstract

Uniformity in Heterogeneity: Diving Deep into Count Interval Partition for Crowd Counting

Cited by 28 publications

References 22 publications

Counting with Adaptive Auxiliary Learning

Counting with Adaptive Auxiliary Learning

Self-supervised Domain Adaptation in Crowd Counting

Occluded Human Mesh Recovery

Contact Info

Product

Resources

About