전력선 인식을 위한 의미론적 세분화 모델과 학습 전략
Semantic Segmentation Model and Training Strategy for Powerline Recognition
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
최근 의미론적 세분화 작업에서 컨볼루션과 트랜스포머 기반의 다양한 딥러닝 모델이 제안되고 있다. 이러한 모델들은 공개 벤치마크의 다양한 객체에 대해 주목할만한 성능을 보여준다. 그러나 우리는 전력선과 같이 전경과 배경의 비율이 과도하게 불균형하며 얇고 긴 객체에 대해 최신 모델조차도 분할에 실패하는 경향이 있음을 관찰한다. 본 논문에서 우리는 전력선 인식을 위해 전력선 데이터 세트를 수집하고, 해당 데이터 세트에서 전력선 검출 실패에 대한 다음의 세 가지 원인이 있음을 발견한다. 1) 레이블링 작업의 어려움, 2) 대규모 훈련 데이터 세트의 부족, 3) 극도로 불균형한 클래스 분포. 더 나아가, 우리는 해당 문제를 해결하기 위해 전력선 인식에 대해 높은 성능을 달성하는 모델 선택에 대한 기준과 훈련 전략을 제시한다. 실험 결과 제시된 학습 전략은 인식 작업에서 기준선 모델과 비교해 20 mIoU 이상의 성능 향상을 달성한다.
Abstract
Recently, various deep learning models based on convolution and transformer have been proposed for semantic segmentation tasks. These models show remarkable performance on a variety of objects on public benchmarks. However, we observe that even state-of-the-art models tend to fail in segmentation for thin and long objects with an excessively imbalanced foreground-background ratio, such as powerlines. In this paper, we collect a powerline dataset for powerline segmentation and identify three main causes for powerline segmentation failure in the dataset. 1) difficulty in labeling tasks, 2) lack of large training datasets, 3) extremely unbalanced class distribution. Furthermore, we present criteria for model selection and training strategies that achieve high performance for powerline segmentation. Our experimental results shows that the presented training strategy achevied an improvement of over 20 mIoU in the powerline segmentation task compared to the baseline model.
참고문헌 (32)
- Patashnik, O., "Styleclip: Text-driven manipulation of stylegan imagery", Proceedings of the IEEE, 2021, 2085~2094
- 김대한, "교차 도메인 혼합 샘플링 기법을 활용한 준 지도 학습 기반 도메인 적응 기법", 한국차세대컴퓨팅학회 논문지, 17(6), pp.28~37 (2021), 한국차세대컴퓨팅학회[DOI]
- Kaymak, Çağrı, "Semantic image segmentation for autonomous driving using fully convolutional networks", International Artificial Intelligence and Data Processing Symposium (IDAP), pp.1~8 (2019)
- Tae-Kyeong Kim, "A Research on the Fainting Monitoring System using Deep Learning", Conference on Information and Control Systems, pp.237~238 (2021)
- 파우델 사하데브, "Medical Image Segmentation via Multi-scale Attention Guided Network", 한국차세대컴퓨팅학회 논문지, 16(3), pp.51~62 (2020), 한국차세대컴퓨팅학회[DOI]
- 챠이트라 다야난다, "A comparative analysis on segmentation of brain MRI using CNN models", 한국차세대컴퓨팅학회 논문지, 16(3), pp.63~76 (2020), 한국차세대컴퓨팅학회[DOI]
- 조다솜, "자동화된 2차원 복부 CT 영상에서 딥러닝 기법을 이용한 근육 분할 기법", 한국차세대컴퓨팅학회 논문지, 18(5), pp.19~27 (2022), 한국차세대컴퓨팅학회[DOI]
- Everingham, M., "The Pascal Visual Object Classes Challenge: A Retrospective", International Journal of Computer Vision, 111, pp.98~136 (2015)
- Vinyals, O., "Show and Tell: Lessons Learned from the 2015 MSCOCO Image Captioning Challenge", IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(4), pp.652~663 (2015)[DOI]
- Cordts, M., "The Cityscapes Dataset for Semantic Urban Scene Understanding", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.3213~3223 (2016)
- O'Shea, K., "An Introduction to Convolutional Neural Networks" (2015)
- Vaswani, A., "Attention is All You Need", Advances in Neural Information Processing Systems, 30 (2017)
- Liu, L., "Deep learning for generic object detection: A survey", International Journal of Computer Vision, 128, pp.261~318 (2020)
- Ericsson, L., "Self-supervised representation learning : Introduction, advances, and challenges", IEEE Signal Processing Magazine, 39(3), pp.42~62 (2022)
- Chen, L. C., "Rethinking atrous convolution for semantic image segmentation" (2017)
- Alexey Dosovitskiy, "An image is worth 16x16 words: Transformers for image recognition at scale", International Conference on Learning Representations (2021)
- Ranftl, R., "Vision transformers for dense prediction", Proceedings of the IEEE, 2021, 12179~12188
- Touvron, H., "Training dataefficient image transformers & distillation through attention", International conference on machine learning, pp.10347~10357 (2021)
- Ze Liu, "Swin transformer: Hierarchical vision transformer using shifted windows", Proceedings of International Conference on Computer Vision (ICCV), pp.10012~10022 (2021)
- Chen Sun, "Revisiting unreasonable effectiveness of data in deep learning era", ICCV (2017)
- Geoffrey E. Hinton, "Distilling the knowledge in a neural network", CoRR (2015)
- Deng, J., "Imagenet: A large-scale hierarchical image database", IEEE conference on computer vision and pattern recognition, pp.248~255 (2009)
- Xiao, T., "Unified perceptual parsing for scene understanding", Proceedings of the European conference on computer vision(ECCV), pp.418~434 (2018)
- Abu Alhaija, H., "Augmented reality meets computer vision : Efficient data generation for urban driving scenes", International Journal of Computer Vision, 126, pp.961~972 (2018)
- MICHAELIS, Claudio, "A Broad Dataset is All You Need for One-Shot Object Detection", ~ (2021)
- Hossain, M. S., "Dual Focal Loss to address class imbalance in semantic segmentation", Neurocomputing, pp.462~ (2021)
- Kim, D., "A broad study of pre-training for domain generalization and adaptation", Computer Vision–ECCV 2022: 17th European Conference, 2022, pp.621~638 (2022)
- Olga Russakovsky, "Imagenet large scale visual recognition challenge", International journal of computer vision, 115(3), pp.211~252 (2015)
- Hu, S. X., "Pushing the limits of simple pipelines for few-shot learning: External data and fine-tuning make a difference", Proceedings of the IEEE, 2022, 9068~9077
- Zhou, B., "Learning deep features for scene recognition using places database", Advances in neural information processing systems, 27 (2014)
- Ilya Loshchilov, "Decoupled weight decay regularization", International Conference on Learning Representations (2019)
- Enze Xie, "Segformer: Simple and efficient design for semantic segmentation with transformers", NeurIPS (2021)