Skip to content

Latest commit

 

History

History
86 lines (69 loc) · 3.54 KB

README.md

File metadata and controls

86 lines (69 loc) · 3.54 KB

2021 BIGDAS Conference

📹 An Accurate Extraction of Facial Meta-Information Using Selective Super Resolution from Crowd Images

Facial Meta-Information Extraction Scheme from Crowd Images

👉 Paper Download

🔎 Why?

  • For crowd monitoring using intelligence video surveillance systems
  • To detect crowd abnormal situations using facial meta-information



⚒ Tech Stacks





📋 Design

image

  1. Face Detection by YOLO5Face
  2. If lower than threshold? (Low Resolution, Not easy)
  3. Super Resolution by ESRGAN to face images
  4. Gender Classification by CNN



🗂 Dataset

  • 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

image

Case Number of Face Images Resolution
Easy Case 596 48x48 ~
Medium Case 5,102 29x29 ~ 48x48
Hard Case 5,432 ~ 29x29



📊 Experiment

1. Face Detection Result

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%

2. Gender Classification Result

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%



👭 Contributions

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



📚 References

  1. Crowd-11 : A Dataset for Fine Grained Crowd Behaviour Analysis, IEEE Conference on CVPRW (2017)
  2. 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)
  3. Delong Qi, Weijun tan, Qi Yao, Jingfeng Liu, YOLO5Face: Why Reinventing a Face Detector, arXiv preprint arXiv:2105.12931 (2021)
  4. Xingtao Wang and others, ESRGAN: Enhanced Super-Resolution Generative Adversarial Network, arXiv preprint arXiv:1809.00219v2 (2018)
  5. Octavio Arriaga and Matias Valdenegro-Toro and Paul Plöger, Real-time Convolutional Neural Networks for Emotion and Gender Classification. arXiv preprint arXiv:1710.07557