Early Stopping Effectiveness for YOLOv4
Downloads
Background: YOLOv4 is one of the fastest algorithms for object detection. Its methods, i.e., bag of freebies and bag of specials, can prevent overfitting, but this can be combined with early stopping as it could also prevent overfitting.
Objective: This study aims to identify the effectiveness of early stopping in preventing overfitting in the YOLOv4 training process.
Methods: Four datasets were grouped based on the training data size and object class, These datasets were tested in the experiment, which was carried out using three patience hyperparameters: 2, 3, and 5. To assess the consistency, it was repeated eight times.
Results: The experimental results show that early stopping is triggered more frequently in training with data below 2,000 images. Of the three patience hyperparameters used, patience 2 and 3 were able to halve the training duration without sacrificing accuracy. Patience 5 rarely triggers early stopping. There is no pattern of correlation between the number of object classes and early stopping.
Conclusion: Early stopping is useful only in training with data below 2,000 images. Patience with a value of 2 or 3 are recommended.
Keywords: Early Stopping, Overfitting, Training data, YOLOv4
X. Ying, "An Overview of Overfitting and its Solutions,” in Journal of Physics: Conference Series, Mar. 2019, vol. 1168, no. 2, doi: 10.1088/1742-6596/1168/2/022022.
Y. Yan, X. Zhang, and B. Wu, "Simplified early stopping criterion for belief-propagation polar code decoder based on frozen bits,” IEEE Access, vol. 7, pp. 134691–134696, 2019, doi: 10.1109/ACCESS.2019.2940135.
A. Effland, E. Kobler, K. Kunisch, and T. Pock, "Variational Networks: An Optimal Control Approach to Early Stopping Variational Methods for Image Restoration,” J. Math. Imaging Vis., vol. 62, no. 3, pp. 396–416, 2020, doi: 10.1007/s10851-019-00926-8.
A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, "YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv, 2020.
A. Bochkovskiy, "AlexeyAB/darknet: YOLOv4v / Scaled-YOLOv4 - Neural Networks for Object Detection (Windows and Linux version of Darknet ),” GitHub, 2020. https://github.com/AlexeyAB/darknet. accesed on 12 Jan2021.
I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT Press, 2016.
D. Jones, E. Nn, and M. L. P. Nn, "Neural Networks,” 2012, pp. 29–35.
D. Kadish, S. Risi, and A. S. Lí¸vlie, "Improving Object Detection in Art Images Using Only Style Transfer,” Feb. 2021, Accessed: May 21, 2021. [Online]. Available: http://arxiv.org/abs/2102.06529.
M. S. Salekin, A. Babaeian Jelodar, and R. Kushol, "Cooking state recognition from images using inception architecture,” 1st Int. Conf. Robot. Electr. Signal Process. Tech. ICREST 2019, pp. 163–168, 2019, doi: 10.1109/ICREST.2019.8644262.
S. Mishra, T. Yamasaki, and H. Imaizumi, "Improving image classifiers for small datasets by learning rate adaptations,” arXiv, 2019.
M. Li, M. Soltanolkotabi, and S. Oymak, "Gradient Descent with Early Stopping is Provably Robust to Label Noise for Overparameterized Neural Networks,” arXiv, Mar. 2019, Accessed: May 21, 2021. [Online]. Available: http://arxiv.org/abs/1903.11680.
S. Chaudhury and T. Yamasaki, "Robustness of Adaptive Neural Network Optimization under Training Noise,” IEEE Access, pp. 37039–37053, 2021, doi: 10.1109/ACCESS.2021.3062990.
J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection,” Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 2016-December, pp. 779–788, Jun. 2015, doi: 10.1109/CVPR.2016.91.
J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger,” Proc. - 30th IEEE Conf. Comput. Vis. Pattern Recognition, CVPR 2017, vol. 2017-January, pp. 6517–6525, Dec. 2016, doi: 10.1109/CVPR.2017.690.
J. Redmon and A. Farhadi, "YOLOv3: An Incremental Improvement,” Apr. 2018, Accessed: Dec. 04, 2021. [Online]. Available: https://arxiv.org/abs/1804.02767v1.
A. Kuznetsova et al., "The Open Images Dataset V4: Unified Image Classification, Object Detection, and Visual Relationship Detection at Scale,” Int. J. Comput. Vis., vol. 128, no. 7, pp. 1956–1981, Jul. 2020, doi: 10.1007/s11263-020-01316-z.
S. Ge, J. Li, Q. Ye, and Z. Luo, "Detecting masked faces in the wild with LLE-CNNs,” in Proceedings - 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Nov. 2017, vol. 2017-Janua, pp. 426–434, doi: 10.1109/CVPR.2017.53.
P. Hidayatullah et al., "DeepSperm: A robust and real-time bull sperm-cell detection in densely populated semen videos,” Comput. Methods Programs Biomed., vol. 209, p. 106302, Sep. 2021, doi: 10.1016/j.cmpb.2021.106302.
R. Padilla, W. L. Passos, T. L. B. Dias, S. L. Netto, and E. A. B. Da Silva, "A comparative analysis of object detection metrics with a companion open-source toolkit,” Electron., vol. 10, no. 3, pp. 1–28, 2021, doi: 10.3390/electronics10030279.
A. Lodwich, Y. Rangoni, and T. Breuel, "Evaluation of robustness and performance of Early Stopping Rules with Multi Layer Perceptrons,” Proc. Int. Jt. Conf. Neural Networks, pp. 1877–1884, 2009, doi: 10.1109/IJCNN.2009.5178626.
Copyright (c) 2022 The Authors. Published by Universitas Airlangga.
This work is licensed under a Creative Commons Attribution 4.0 International License.
Authors who publish with this journal agree to the following terms:
All accepted papers will be published under a Creative Commons Attribution 4.0 International (CC BY 4.0) License. Authors retain copyright and grant the journal right of first publication. CC-BY Licenced means lets others to Share (copy and redistribute the material in any medium or format) and Adapt (remix, transform, and build upon the material for any purpose, even commercially).
