☆ 4.3 Article

Deformable channel non-local network for crowd counting

ELECTRONICS LETTERS (2023)

NSSNet: Scale-Aware Object Counting With Non-Scale Suppression

Liang Liu et al.

Summary: In the field of object counting, it is important to consider the variation in object sizes at different scales, especially in crowd and vehicle counting for intelligent transportation. This paper proposes a scale-aware counting network called NSSNet, which generates multi-scale predictions and applies a novel non-scale suppression operator to eliminate mismatched estimations. The performance evaluation on standard benchmarks demonstrates the state-of-the-art performance of NSSNet. The scale adaptability of NSSNet is also validated through controlled multi-scale experiments.

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS (2022)

添加到收藏夹

Article Automation & Control Systems

Feature-Aware Adaptation and Density Alignment for Crowd Counting in Video Surveillance

Junyu Gao et al.

Summary: With the advancement of deep neural networks, the performance of crowd counting and pixel-wise density estimation has been improving, but challenges persist. A synthetic crowd dataset released recently helps address the difficulty in collecting data, but the domain gap between real data and synthetic images hinders model performance. To tackle this, a domain-adaptation-style crowd counting method is proposed in this article, which effectively adapts the model from synthetic data to specific real-world scenes.

IEEE TRANSACTIONS ON CYBERNETICS (2021)

添加到收藏夹

Article Engineering, Electrical & Electronic