Learning From Synthetic Data for Crowd Counting in the Wild

Qi Wang, Junyu Gao, Yuan Yuan

2019 · 604 citations · 30 references

Concepts

TL;DR

Crowd counting is a hot topic for applications such as video surveillance and public security, but it remains challenging in the wild due to variable environments, large crowd ranges, and limited data that cause over‑fitting. The study aims to address data scarcity and performance issues by creating a synthetic crowd dataset and proposing two methods—pretraining with synthetic data followed by fine‑tuning, and a domain‑adaptation counting approach—to improve crowd counting in the wild. The authors build a large‑scale synthetic dataset using an automated collector and labeler, then train a crowd counter on it and fine‑tune with real data, and also develop a domain‑adaptation counting method that reduces the need for manual annotations. Experiments demonstrate that pretraining on synthetic data followed by fine‑tuning achieves state‑of‑the‑art results on four real datasets, while the domain‑adaptation method surpasses baseline approaches; the dataset and code are publicly released.

Abstract

Recently, counting the number of people for crowd scenes is a hot topic because of its widespread applications (e.g. video surveillance, public security). It is a difficult task in the wild: changeable environment, large-range number of people cause the current methods can not work well. In addition, due to the scarce data, many methods suffer from over-fitting to a different extent. To remedy the above two problems, firstly, we develop a data collector and labeler, which can generate the synthetic crowd scenes and simultaneously annotate them without any manpower. Based on it, we build a large-scale, diverse synthetic dataset. Secondly, we propose two schemes that exploit the synthetic data to boost the performance of crowd counting in the wild: 1) pretrain a crowd counter on the synthetic data, then finetune it using the real data, which significantly prompts the model's performance on real data; 2) propose a crowd counting method via domain adaptation, which can free humans from heavy data annotations. Extensive experiments show that the first method achieves the state-of-the-art performance on four real datasets, and the second outperforms our baselines. The dataset and source code are available at https://gjy3035.github.io/GCC-CL/.

References

30