Physics-guided Data Augmentation for Learning the Solution Operator of Linear Differential Equations

Li, Ye; Pang, Yiwen; Shu, Bin

doi:10.1109/ccis57298.2022.10016380

Cited by 1 publication

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast to existing state-of-the-art methods [8], [61], [29], [50], [57] and [62] our proposed framework harnesses the full potential of intra and inter-depth information representation through innovative convolutional modules, specifically employing a nonincreasing-order (NIO) kernel arrangement. To assess the individual contributions of each component within our framework, we introduce a non-static kernel arrangement convolutional baseline that employs truncation and relies on a smart network with the same scene projection approach.…”

Section: Effectiveness Of Convolutional Arrangementmentioning

confidence: 99%

“…• Feature Augmentation: Image augmentation techniques are commonly used in computer vision tasks to improve the model generalization [62] and increase the diversity of the training datasets. This work leverages some well-known data augmentation techniques such as Rotation, Horizontal Flip, Random Crop, Brightness and Contrast Adjustment, Gaussian Noise, Color Jitter Elastic Deformation [63], and cut-out [64], [55] to make simulation work more impactful for confident head feature selection and extraction in images.…”

Section: Problem Formulationmentioning

confidence: 99%

See 1 more Smart Citation

ED-DKCNN: Advancing Supervised Crowd Counting in Complex Environments

Tomar,

Kumar,

Nijhawan

et al. 2024

Preprint

View full text Add to dashboard Cite

The increasing urban population has led to challenges in managing crowd dynamics, especially preventing tragic incidents like stampedes. Real-time, accurate crowd counting faces obstacles such as background clutter and perspective variations. The study accepts these challenges of supervised crowd counting by examining the effectiveness of convolutional arrangements in improving accuracy by using an encoder-decoder dynamic convolutional neural network * (ED-DKCNN). The combined segmented, edge-oriented data and texture-rich features are the input for the model capable of precise crowd counting, achievable even in complex scenarios with occlusions and dense crowds. It explores low and high-level crowd features, addresses occlusion and uneven crowd distribution, and utilizes deep mining and dense complementarity for optimized people counting without density map estimation. The proposed framework harnesses intra-and inter-depth information representation through a non-increasing-order kernel arrangement, achieving state-of-the-art accuracy in people counting compared to existing methods across various datasets. The extensive experiments over free and surveillance category datasets through multiple evaluation criteria firmly * https://link.springer.com/article/10. establish the proposed ED-DKCNN model as a state-of-the-art performer in this domain. Moreover, the proposed model significantly advances crowd-counting methodologies, offering potential applications in multi-modal data integration, real-time scenarios, privacy concerns, edge computing, cross-domain situations, and human behavior.

show abstract