📹 An Accurate Extraction of Facial Meta-Information Using Selective Super Resolution from Crowd Images
- For crowd monitoring using intelligence video surveillance systems
- To detect crowd abnormal situations using facial meta-information
- Face Detection by YOLO5Face
- If lower than threshold? (Low Resolution, Not easy)
- Super Resolution by ESRGAN to face images
- Gender Classification by CNN
- Crowd Dataset in real world
- Extract total of 551 images by sampling 111 videos from Crowd-11 Dataset
- Get annotation files by labeling the x and y coordinate values of face region and gender of the face
- 6,146 male faces and 4,984 female faces in total 11,130 faces
- 👉 Dataset Download
| Case | Number of Face Images | Resolution |
|---|---|---|
| Easy Case | 596 | 48x48 ~ |
| Medium Case | 5,102 | 29x29 ~ 48x48 |
| Hard Case | 5,432 | ~ 29x29 |
| Case | Recall | Accuracy |
|---|---|---|
| Easy Case | 87.67% | 84.41% |
| Medium Case | 85.04% | 80.32% |
| Hard Case | 74.79% | 59.99% |
| Total | 74.79% | 71.1% |
| Case | Before ESRGAN | After ESRGAN |
|---|---|---|
| Easy Case | 72.15% | - |
| Medium Case | 68.86% | 69.78% |
| Hard Case | 65.24% | 65.24% |
| Total | 67.27% | 68.20% |
| Name | Contribution | Contact |
|---|---|---|
| Jieun Park | Super Resolution, Gender Classification, Dataset, Author | jieunpark@inha.edu |
| Yurim Kang | Face Detection, Dataset | 12171745@inha.edu |
| Yoosung Kim | Corresponding Author | yskim@inha.ac.kr |
- Crowd-11 : A Dataset for Fine Grained Crowd Behaviour Analysis, IEEE Conference on CVPRW (2017)
- Yang, Shuo and Luo, Ping and Loy, Chen change and Tang, Xiaoou, WIDER FACE: A Face Detection Benchmark, IEEE conference on Computer Vision and Pattern Recognition (CVPR) (2016)
- Delong Qi, Weijun tan, Qi Yao, Jingfeng Liu, YOLO5Face: Why Reinventing a Face Detector, arXiv preprint arXiv:2105.12931 (2021)
- Xingtao Wang and others, ESRGAN: Enhanced Super-Resolution Generative Adversarial Network, arXiv preprint arXiv:1809.00219v2 (2018)
- Octavio Arriaga and Matias Valdenegro-Toro and Paul Plöger, Real-time Convolutional Neural Networks for Emotion and Gender Classification. arXiv preprint arXiv:1710.07557