面向目标检测的数据集标签遗漏的协同修正技术

doi:10.3778/j.issn.1002-8331.2302-0056

摘要/Abstract

摘要： 针对图像标注中疲劳、粗心等因素引起的标签遗漏现象，使得模型训练时难以正确区分正负样本，进而影响模型性能。设计了一种协同修正技术，通过多次迭代更新训练集，将潜在无标签的目标进行对象擦除，降低训练集的错误监督信息，避免人工的重复检查和重复标注。该方法无需进行算法参数调整、不依赖具体网络结构，低成本地减少数据集错误从而提升模型训练精度。在YOLOv5算法的实验基础上表明协同修正操作仅迭代1次即有明显的改善效果，并在多个公共数据集上能够提升0.4%~1.4%的检测精度，当数据集中的标签遗漏率达到40%时依然能够生效。该方法对数据集中样本的数据量和类别数没有限制，可应用于电商、遥感、通用等多种目标检测场景，保持着较好的鲁棒性和泛化性。

关键词: 协同修正, 标签遗漏, 数据集优化, 目标检测, 深度学习

Abstract: For the label omission caused by fatigue, carelessness and other factors in image labeling, it is difficult to correctly distinguish positive and negative samples during model training, thus affecting the performance of the model. A collaborative correction technology is designed to update the training set through multiple rounds of iteration, erase the potential unlabeled object, reduce the error monitoring information of the training set, and avoid manual repeated inspection and labeling. This method does not need to adjust the algorithm parameters, does not depend on the specific network structure, and reduces the dataset errors at low cost to improve the model training accuracy. Based on the experiment of YOLOv5 algorithm, it is shown that the cooperative correction operation can improve the detection accuracy by 0.4%~1.4% on multiple common datasets after only one iteration, and it still takes effect when the label omission rate in the dataset reaches 40%. This method has no limit on the amount of data and the number of categories of samples in the dataset, and can be applied to multiple target detection scenarios such as e-commerce, remote sensing, general purpose, etc., maintaining good robustness and generalization.

Key words: collaborative correction, label omission, dataset optimization, object detection, deep learning

周定威, 扈静, 张良锐, 段飞亚. 面向目标检测的数据集标签遗漏的协同修正技术[J]. 计算机工程与应用, 2024, 60(8): 267-273.

ZHOU Dingwei, HU Jing, ZHANG Liangrui, DUAN Feiya. Collaborative Correction Technology of Label Omission in Dataset for Object Detection[J]. Computer Engineering and Applications, 2024, 60(8): 267-273.

参考文献

[1] SUN C, SHRIVASTAVA A, SINGH S, et al. Revisiting unreasonable effectiveness of data in deep learning era[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 843-852.
[2] WANG X, HUANG T E, DARRELL T, et al. Frustratingly simple few-shot object detection[C]//Proceedings of the 37th International Conference on Machine Learning, 2020: 9919-9928.
[3] NORTHCUTT C G, ATHALYE A, MUELLER J. Pervasive label errors in test sets destabilize machine learning benchmarks[J]. arXiv:2103.14749, 2021.
[4] ZHANG H, CHEN F, SHEN Z, et al. Solving missing-annotation object detection with background recalibration loss[C]//Proceedings of the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing, 2020: 1888-1892.
[5] ZHANG Y, CHENG Y, HUANG X, et al. Simple and robust loss design for multi-label learning with missing labels[J]. arXiv:2112.07368, 2021.
[6] CHENG Y, QIAN K, MIN F. Global and local attention-based multi-label learning with missing labels[J].Information Sciences, 2022, 594: 20-42.
[7] TAN A, LIANG J, WU W Z, et al. Semi-supervised partial multi-label classification via consistency learning[J].Pattern Recognition, 2022, 131: 108839.
[8] WANG T, YANG T, CAO J, et al. Co-mining: self-supervised learning for sparsely annotated object detection[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2021: 2800-2808.
[9] GOLDBERGER J, BENREUVEN E. Training deep neural-networks using a noise adaptation layer[C]//Proceedings of the International Conference on Learning Representations, 2016.
[10] JINDAL I, NOKLEBY M, CHEN X. Learning deep networks from noisy labels with dropout regularization[C]//Proceedings of the 2016 IEEE 16th International Conference on Data Mining, 2016: 967-972.
[11] SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15（1）: 1929-1958.
[12] GOODFELLOW I J, SHLENS J, SZEGEDY C. Explaining and harnessing adversarial examples[J].arXiv:1412.6572, 2014.
[13] MIYATO T, DAI A M, GOODFELLOW I. Adversarial training methods for semi-supervised text classification[J]. arXiv:1605.07725, 2021.
[14] MADRY A, MAKELOV A, SCHMIDT L, et al. Towards deep learning models resistant to adversarial attacks[J]. arXiv:1706.06083, 2019.
[15] SHAFAHI A, NAJIBI M, GHIASI M A, et al. Adversarial training for free![C]//Proceedings of the 33rd International Conference on Neural Information Processing Systems, 2019: 3358-3369.
[16] ZHANG D, ZHANG T, LU Y, et al. You only propagate once: accelerating adversarial training via maximal principle[C]//Proceedings of the 33rd International Conference on Neural Information Processing Systems, 2019: 227-238.
[17] ZHU C, CHENG Y, GAN Z, et al. FreeLB: enhanced adversarial training for natural language understanding[J].arXiv:1909.11764, 2019.
[18] MIYATO T, MAEDA S, KOYAMA M, et al. Virtual adversarial training: a regularization method for supervised and semi-supervised learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 41（8）: 1979-1993.
[19] DEVRIES T, TAYLOR G W. Improved regularization of convolutional neural networks with cutout[J]. arXiv:1708. 04552, 2017.
[20] GE Z, LIU S, WANG F, et al. YOLOX: exceeding YOLO series in 2021[J]. arXiv:2107.08430, 2021.
[21] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39（6）: 1137-1149.