Bi-level Alignment for Cross-Domain Crowd Counting

Recently, crowd density estimation has received increasing attention. The main challenge for this task is to achieve high-quality manual annotations on a large amount of training data. To avoid reliance on such annotations, previous works apply unsupervised domain adaptation (UDA) techniques by transferring knowledge learned from easily accessible synthetic data to real-world datasets. However, current state-of-the-art methods either rely on external data for training an auxiliary task or apply an expensive coarse-to-fine estimation. In this work, we aim to develop a new adversarial learning based method, which is simple and efficient to apply. To reduce the domain gap between the synthetic and real data, we design a bi-level alignment framework (BLA) consisting of (1) task-driven data alignment and (2) fine-grained feature alignment. In contrast to previous domain augmentation methods, we introduce AutoML to search for an optimal transform on source, which well serves for the downstream task. On the other hand, we do fine-grained alignment for foreground and background separately to alleviate the alignment difficulty. We evaluate our approach on five real-world crowd counting benchmarks, where we outperform existing approaches by a large margin. Also, our approach is simple, easy to implement and efficient to apply. The code is publicly available at https://github.com/Yankeegsj/BLA.

Automated summary

Recently, there has been increasing attention on crowd density estimation. The main challenge lies in obtaining high-quality manual annotations on a large amount of training data. To overcome this reliance on annotations, previous works have applied unsupervised domain adaptation (UDA) techniques to transfer knowledge from synthetic data to real-world datasets. However, current state-of-the-art methods either rely on external data for auxiliary task training or use expensive coarse-to-fine estimation. In this work, we propose a novel adversarial learning based method that is simple and efficient. To reduce the domain gap between synthetic and real data, we introduce a bi-level alignment framework consisting of task-driven data alignment and fine-grained feature alignment. Unlike previous domain augmentation methods, we incorporate AutoML to search for an optimal transform on the source data for improved downstream task performance. Additionally, we perform fine-grained alignment for foreground and background separately to alleviate alignment difficulties. Our approach outperforms existing methods by a large margin on five real-world crowd counting benchmarks. It is also easy to implement and efficient. The code is publicly available at https://github.com/Yankeegsj/BLA.