Feature-aware Adaptation and Density Alignment for Crowd Counting in Video Surveillance
This addresses the challenge of generalizing crowd counting models to unseen real-world domains, which is incremental as it builds on existing domain adaptation techniques.
The paper tackles the problem of domain gap between synthetic and real-world data in crowd counting by proposing a domain adaptation method with multi-level feature adaptation and structured density alignment, achieving state-of-the-art performance on four surveillance datasets.
With the development of deep neural networks, the performance of crowd counting and pixel-wise density estimation are continually being refreshed. Despite this, there are still two challenging problems in this field: 1) current supervised learning needs a large amount of training data, but collecting and annotating them is difficult; 2) existing methods can not generalize well to the unseen domain. A recently released synthetic crowd dataset alleviates these two problems. However, the domain gap between the real-world data and synthetic images decreases the models' performance. To reduce the gap, in this paper, we propose a domain-adaptation-style crowd counting method, which can effectively adapt the model from synthetic data to the specific real-world scenes. It consists of Multi-level Featureaware Adaptation (MFA) and Structured Density map Alignment (SDA). To be specific, MFA boosts the model to extract domain-invariant features from multiple layers. SDA guarantees the network outputs fine density maps with a reasonable distribution on the real domain. Finally, we evaluate the proposed method on four mainstream surveillance crowd datasets, Shanghai Tech Part B, WorldExpo'10, Mall and UCSD. Extensive experiments evidence that our approach outperforms the state-of-the-art methods for the same cross-domain counting problem.