GCC Dataset
GCC dataset consists of 15,212 images, with resolution of 1080×1920, containing 7,625,843 persons. Compared with the existing datasets, GCC is a more large-scale crowd counting dataset in both the number of images and the number of persons.
Figure 1. The exemplars of synthetic crowd scenes from the proposed GCC dataset.
Table 1 shows the basic information of GCC and the existing datasets. In addition to the above advantages, GCC is more diverse than other real-world datasets.
| Dataset | Number of Images | Average Resolution | Count Statistics | |||
|---|---|---|---|---|---|---|
| Total | Min | Age | Max | |||
| UCSD | 2,000 | 158 × 238 | 49,885 | 11 | 25 | 46 |
| Mall | 2,000 | 480 × 640 | 62,325 | 13 | 31 | 53 |
| UCF_CC_50 | 50 | 2101 × 2888 | 63,974 | 94 | 1,279 | 4,543 |
| WorldExpo10 | 3,980 | 576 × 720 | 199,932 | 1 | 50 | 253 |
| SHT A | 482 | 589 × 868 | 241,677 | 33 | 501 | 3,139 |
| SHT B | 716 | 768 × 1024 | 88,488 | 9 | 123 | 578 |
| UCF-QNRF | 1,525 | 2013 × 2902 | 1,251,642 | 49 | 815 | 12,865 |
| GCC | 15,212 | 1080 × 1920 | 7,625,843 | 0 | 501 | 3995 |
Table 1. Statistics of the seven real-world datasets and the synthetic GCC dataset.
Deep Learning Model for Crowd Counting
Supervised Crowd Counting
We present a pretrained scheme to prompt the original method's performance on the real data, which effectively reduces the estimation errors compared with random initialization and ImageNet model, respectively. Further, through the strategy, our proposed SFCN achieves the state-of-the-art results.
Crowd Counting via Domain Adaptation
We propose a crowd counting method via domain adaptation, which can effectively learn domain-invariant feature between synthetic and real data. To be specific, we present a SSIM Embedding (SE) Cycle GAN to transform the synthetic image to the photo-realistic image. Then we will train a SFCN on the translated data. Finally, we directly test the model on the real data.
Download Place
Bibtex
Please cite the following paper if you use our work.
@inproceedings{wang2019learning,
title={Learning from Synthetic Data for Crowd Counting in the Wild},
author={Wang, Qi and Gao, Junyu and Lin, Wei and Yuan, Yuan},
booktitle={Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
pages={8198--8207},
year={2019}
}