Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds

Idrees, Haroon; Tayyab, Muhmmad; Athrey, Kishan; Zhang, Dong; Al-Maadeed, Somaya; Rajpoot, Nasir M.; Shah, Mubarak

doi:10.48550/arxiv.1808.01050

Cited by 17 publications

(67 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…UCF-QNRF. The UCF-QNRF dataset was recently introduced by [21]. It is currently the largest crowd dataset Method MAE RMSE Zhang et al [64] 1.60 3.31 CCNN [43] 1.51 -Switch-CNN [50] 1.62 2.10 FCN-rLSTM [65] 1.54 3.02 CSRNet [32] 1.16 1.47 MCNN [66] 1.07 1.35 DUB-CSRNet (Ours) 1.03 1.24 SANet [5] 1.02 1.29 [19] 315 508 MCNN [66] 277 426 Encoder-Decoder [2] 270 478 CTML [56] 252 514 Switch-CNN [50] 228 445 Resnet101 [17] 190 277 Densenet201 [18] 163 226 Idrees et al(2018) [21] 132 191 DUB-CSRNet (Ours) 116 178…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Crowd Counting with Decomposed Uncertainty

Oh¹,

Olsen²,

Ramamurthy³

2019

Preprint

View full text Add to dashboard Cite

Research in neural networks in the field of computer vision has achieved remarkable accuracy for point estimation. However, the uncertainty in the estimation is rarely addressed. Uncertainty quantification accompanied by point estimation can lead to a more informed decision, and even improve the prediction quality. In this work, we focus on uncertainty estimation in the domain of crowd counting. We propose a scalable neural network framework with quantification of decomposed uncertainty using a bootstrap ensemble. We demonstrate that the proposed uncertainty quantification method provides additional insight to the crowd counting problem and is simple to implement. We also show that our proposed method outperforms the current state of the art method in many benchmark datasets. To the best of our knowledge, we have the one of the best systems for ShanghaiTech part A and B, UCF-CC 50, UCSD, and the best for UCF-QNRF datasets.

show abstract

Section: Methodsmentioning

confidence: 99%

“…We evaluate our method on four publicly available crowd counting datasets: ShanghaiTech [66], UCF-CC 50 [20], UCSD [6], and UCF-QNRF [21]. For all datasets, we generate ground truth density maps with fixed spread Gaussian kernel (see Section C in the appendix for details).…”

Section: Performance Comparisonsmentioning

confidence: 99%

Crowd Counting with Decomposed Uncertainty

Oh¹,

Olsen²,

Ramamurthy³

2019

Preprint

View full text Add to dashboard Cite

show abstract

“…This section provides data preprocessing strategies of six mainstream datasets. They are UCF CC 50 [4], World-Expo '10 [12], SHT A [13], SHT B [13], UCF-QNRF [5], and GCC [11], as shown in Table 2. Overall, data preprocessing strategies are constituted by mainly two parts, including the input size and the transformation about ground truth.…”

Section: Data Preprocessing Strategymentioning

confidence: 99%

“…SHT A [13] geometry-adaptive kernels keep the original height-width ratio, max(h, w) = 1024, min(h, w)%16 = 0 SHT B [13] 15 × 15 original size: 768 × 1024 WE [12] 15 × 15 original size: 576 × 720 QNRF [5] 15 × 15 keep the original height-width ratio, max(h, w) = 1024, min(h, w)%16 = 0 GCC [11] 15 × 15 resize to 544 × 960 batch size for those pre-trained models (Alexnet, VGG, ResNet, etc. ), and multiple batch size for models trained from scratch.…”

Section: Input Sizementioning

confidence: 99%

“…Recently, crowd counting has made overwhelming development with the rise of deep learning. On one hand, many large-scale datasets with human annotations are published in these years, e.g., UCF CC 50 [4], ShangHaiTech part A and B [13], UCF-QNRF [5] and GCC [11]. one the other hand, many CNN-based models are developed, e.g., MCNN [13], CSRNet [7], SANet [2].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

C^3 Framework: An Open-source PyTorch Code for Crowd Counting

Ji¹,

Lin²,

Zhao³

et al. 2019

Preprint

View full text Add to dashboard Cite

This technical report attempts to provide efficient and solid kits addressed on the field of crowd counting, which is denoted as Crowd Counting Code Framework (C 3 F). The contributions of C 3 F are in three folds: 1) Some solid baseline networks are presented, which have achieved the stateof-the-arts. 2) Some flexible parameter setting strategies are provided to further promote the performance. 3) A powerful log system is developed to record the experiment process, which can enhance the reproducibility of each experiment. Our code is made publicly available at https:// github.com/gjy3035/C-3-Framework. Furthermore, we also post a Chinese blog 1 to describe the details and insights of crowd counting.

show abstract

Performance Comparison of Deep Residual Networks-Based Super Resolution Algorithms Using Thermal Images: Case Study of Crowd Counting

Rizvi

Farooq

Raza

2022

Digital Interaction and Machine Intelligence

View full text Add to dashboard Cite

Humans are able to perceive objects only in the visible spectrum range which limits the perception abilities in poor weather or low illumination conditions. The limitations are usually handled through technological advancements in thermographic imaging. However, thermal cameras have poor spatial resolutions compared to RGB cameras. Super-resolution (SR) techniques are commonly used to improve the overall quality of low-resolution images. There has been a major shift of research among the Computer Vision researchers towards SR techniques particularly aimed for thermal images. This paper analyzes the performance of three deep learning-based state-of-the-art SR algorithms namely Enhanced Deep Super Resolution (EDSR), Residual Channel Attention Network (RCAN) and Residual Dense Network (RDN) on thermal images. The algorithms were trained from scratch for different upscaling factors of ×2 and ×4. The dataset was generated from two different thermal imaging sequences of BU-TIV benchmark. The sequences contain both sparse and highly dense type of crowds with a far field camera view. The trained models were then used to super-resolve unseen test images. The quantitative analysis of the test images was performed using common image quality metrics such as PSNR, SSIM and LPIPS, while qualitative analysis was provided by evaluating effectiveness of the algorithms for crowd counting application. After only 54 and 51 epochs of RCAN and RDN respectively, both approaches were able to output average scores of 37.878, 0.986, 0.0098 and 30.175, 0.945, 0.0636 for PSNR, SSIM and LPIPS respectively. The EDSR algorithm took the least computation time during both training and testing because of its simple architecture. This research proves that a reasonable accuracy can be achieved with fewer training epochs when an application-specific dataset is carefully selected.

show abstract

Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds

Cited by 17 publications

References 0 publications

Crowd Counting with Decomposed Uncertainty

Crowd Counting with Decomposed Uncertainty

C^3 Framework: An Open-source PyTorch Code for Crowd Counting

Performance Comparison of Deep Residual Networks-Based Super Resolution Algorithms Using Thermal Images: Case Study of Crowd Counting

Contact Info

Product

Resources

About