CNN-Based cascaded multi-task learning of high-level prior and density estimation for crowd counting

By Vishwanath A. Sindagi, Vishal M. Patel. Advanced Video and Signal Based Surveillance, 2017

537 citations recorded by the source corpus.

Estimating crowd count in densely crowded scenes is an extremely challenging task due to non-uniform scale variations. In this paper, we propose a novel end-to-end cascaded network of CNNs to jointly learn crowd count classification and density map estimation. Classifying crowd count into various groups is tantamount to coarsely estimating the total count in the image thereby incorporating a high-level prior into the density estimation network. This enables the layers in the network to learn globally relevant discriminative features which aid in estimating highly refined density maps with lower count error. The joint training is performed in an end-to-end fashion. Extensive experiments on highly challenging publicly available datasets show that the proposed method achieves lower count error and better quality density maps as compared to the recent state-of-the-art methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/9fb7a23910f6464902f1b653025f3aeaa20b90dd

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes (2018, methodology): “[39] propose a CNN which uses the high-level prior information to boost the density prediction performance.”
Bayesian Loss for Crowd Count Estimation With Point Supervision (2019, unknown): “1 CMTL [40] 252 514 101.”
Learning From Synthetic Data for Crowd Counting in the Wild (2019, methodology): “[42, 35] exploit multi-task learning to explore Table 1.”
Context-Aware Crowd Counting (2018, unknown): “[14] 315 508 MCNN [42] 277 426 Encoder-Decoder [3] 270 478 CMTL [35] 252 514 Switch-CNN [31] 228 445 Resnet101 [12] 190 277 Densenet201 [13] 163 226 Idrees et al.”
Generating High-Quality Crowd Density Maps Using Contextual Pyramid CNNs (2017, methodology): “[32] proposed to learn a high-level prior and perform density estimation in a cascaded setting.”
Attention Scaling for Crowd Counting (2020, methodology): “These methods have tried to exploit multi-scale feature fusion [49, 29, 35], multi-task learning [34, 30, 21], and the attention mechanism [45, 44, 19] to solve the above questions.”
Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds (2018, methodology): “For counting, we evaluated the new UCF-QNRF dataset using the proposed method which estimates counts, density maps and locations of people simultaneously with several state-of-the-art deep neural networks [3], [8], [10] as well as those specifically developed for crowd counting [30], [25], [24].”
JHU-CROWD++: Large-Scale Crowd Counting Dataset and A Benchmark Method (2020, methodology): “Specifically, we evaluate the following recent works: mulit-column network (MCNN) [5], cascaded multi-task learning for crowd counting (CMTL) [24], CSR-Net [75], SA-Net [26], context-aware crowd counting (CACC) [76], spatial fully convolutional network (SFCN) [44], deep structured scale integration ”
Scale Aggregation Network for Accurate and Efficient Crowd Counting (2018, methodology): “(2) [5,32,6] require density level classifier to provide contextual information.”
ADCrowdNet: An Attention-Injective Deformable Convolutional Network for Crowd Understanding (2018, methodology): “We first compare the variants of the proposed ADCrowdNet network with the state-of-art work CSRNet [14] along with several previous methods including CP-CNN [24], MCNN [32], Cascaded-MTL [23], Switching-CNN [20] on the ShanghaiTech dataset and the UCF CC 50 dataset.”
Crowd Counting via Adversarial Cross-Scale Consistency Pursuit (2018, unknown): “[34] 467.0 498.5 MCNN [37] 377.6 509.1 Cascaded-MTL [27] 322.8 341.4 Switch-CNN [25] 318.1 439.2 CP-CNN [28] 295.8 320.9 ACSCP (ours) 291.0 404.6 Table 5.”