산업 환경의 객체 검출 성능 개선을 위한 GAN 기반 이미지 데이터 증강
GAN-based Image Data Augmentation for Improving Object Detection Performance in Industrial Environments
0
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
객체 검출은 예상치 못한 위험 상황을 회피할 수 있도록 작업자에게 자동으로 경보를 제공할 수 있는 중요한 산업 안전 기술 중 하나이다. 그러나 딥 러닝 기반 객체 감지 모델은 더 높은 성능을 달성하기 위해 대규모의 학습 데이터가 요구되며, 데이터 수집 및 레이블링 작업은 인력이 필요한 힘든 작업이다. 이러한 한계를 해결하기 위해, 본 논문에서는 본래의 데이터세트를 더욱 다양한 데이터로 보충할 수 있는 GAN 기반 데이터 증강을 제안한다. 또한, 생성된 데이터의 충실도를 향상시키기 위한 트랜스포머 기반 생성자 네트워크를제시하며, 비교 연구를 위해 다양한 증강 설정에서 훈련된 기존 객체 감지 모델(YOLOv5)을 평가한다. 평가결과는 20% 증강된 데이터로 학습한 모델이 객체 특정 성능의 손실 없이 0.9%의 분류 능력이 향상되었음을보여준다.
Abstract
Object detection is one of the important industrial safety technologies that can automatically provide a worker with alerts to avoid unexpected near misses. However, deep learning-based object detection models require large amounts of training data to achieve higher performance, and data collection and labeling work is laborious and requires human resources. To address these limitations, we propose a GAN-based data augmentation that can supplement the original dataset with more diverse examples. In addition, we present a transformer-based generator network to improve the fidelity of generated data and evaluate the existing object detection model(YOLOv5) trained under different augmentation settings for a comparison study. The evaluation results show that the classification ability of the model trained with 20% augmented data has improved by 0.9% without localization performance losses.
참고문헌 (26)
- Deng Benyang, "Safety helmet detection method based on YOLO v4", 2020 16th International Conference on Computational Intelligence and Security (CIS), pp.155~158 (2020), IEEE
- Nipun D. Nath, "Deep learning for site safety: Real-time detection of personal protective equipment", Automation in Construction, 112, pp.103085~ (2020), Elsevier BV[DOI]
- Congcong Zhao, "Real-Time Pedestrian Detection Based on Improved YOLO Model", 2019 11th International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC), 2 (2019), IEEE
- He Xu, "Vehicle and Pedestrian Detection Algorithm Based on Lightweight YOLOv3-Promote and Semi-Precision Acceleration", IEEE Transactions on Intelligent Transportation Systems, 23(10), pp.19760~19771 (2022), Institute of Electrical and Electronics Engineers (IEEE)[DOI]
- Alon Halevy, "The Unreasonable Effectiveness of Data", IEEE Intelligent Systems, 24(2), pp.8~12 (2009), Institute of Electrical and Electronics Engineers (IEEE)[DOI]
- Sun C., "Revisiting unreasonable effectiveness of data in deep learning era", Proceedings of the IEEE international conference on computer vision (2017)
- Muhammad Ahmed, "Survey and Performance Analysis of Deep Learning Based Object Detection in Challenging Environments", Sensors, 21(15), pp.5116~ (2021), MDPI AG[DOI]
- "ImageNet Large Scale Visual Recognition Challenge (ILSVRC)"[URL]
- Ian Goodfellow, "Generative adversarial networks", Communications of the ACM, 63(11), pp.139~144 (2020), Association for Computing Machinery (ACM)[DOI]
- Karras T, "Analyzing and improving the image quality of stylegan", Proceedings of the IEEE, 2020, 8110~8119
- Zhao L, "Improved transformer for high-resolution gans", Advances in Neural Information Processing Systems, 34, pp.18367~18380 (2021)
- Maayan Frid-Adar, "GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification", Neurocomputing, 321, pp.321~331 (2018), Elsevier BV[DOI]
- Hojjat Salehinejad, "Generalization of Deep Neural Networks for Chest Pathology Classification in X-Rays Using Generative Adversarial Networks", 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.90~994 (2018), IEEE
- Chenglong Wang, "Lychee Surface Defect Detection Based on Deep Convolutional Neural Networks with GAN-Based Data Augmentation", Agronomy, 11(8), pp.1500~ (2021), MDPI AG[DOI]
- Jiang Y., "Transgan: Two transformers can make one strong gan", arXiv preprint arXiv:2102.07074, ~ (2021)
- Zhao S., "Differentiable augmentation for data-efficient gan training", Advances in Neural Information Processing Systems, 33, pp.7559~7570 (2020)[DOI]
- Bokun Liu, "The Effect of Data Augmentation Methods on Pedestrian Object Detection", Electronics, 11(19), pp.3185~ (2022), MDPI AG[DOI]
- Inoue H., "Data augmentation by pairing samples for images classification", arXiv preprint arXiv:1801.02929, ~ (2018)
- He K., "Deep residual learning for image recognition", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.770~778 (2016)
- "YOLOv5(Ultralytics)"[URL]
- Dosovitskiy A., "An image is worth 16x16 words: Transformers for image recognition at scale", arXiv preprint arXiv:2010.11929, ~ (2020)
- Ali Borji, "Pros and cons of GAN evaluation measures", Computer Vision and Image Understanding, 179, pp.41~65 (2019), Elsevier BV[DOI]
- Heusel M., "Gans trained by a two time-scale update rule converge to a local nash equilibrium", Advances in neural information processing systems, 30 (2017)[DOI]
- Christian Szegedy, "Rethinking the Inception Architecture for Computer Vision", 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp.2818~2826 (2016), IEEE
- Nobuyuki Otsu, "A Threshold Selection Method from Gray-Level Histograms", IEEE Transactions on Systems, Man, and Cybernetics, 9(1), pp.62~66 (1979), Institute of Electrical and Electronics Engineers (IEEE)[DOI]
- Lin T. Y., "Microsoft coco: Common objects in context", European conference on computer vision, pp.740~755 (2014), Springer