Calibrating Fake Feature Statistical Distribution for Few-Shot Object Detection

Lanlan Liu; Haifeng Sima

doi:10.62051/ijcsit.v7n3.04

Authors

Lanlan Liu
Haifeng Sima

DOI:

https://doi.org/10.62051/ijcsit.v7n3.04

Keywords:

Meta-learning, Data augmentation, Distribution calibration, Generative adversarial network

Abstract

The booming development of object detection has benefited from rich datasets. However, obtaining a large amount of labeled data for scarce objects is expensive. Few-Shot Object Detection (FSOD) has garnered significant attention, aiming to learn new categories with only a small number of labeled samples. In this paper, we propose a novel FSOD model based on data augmentation, namely Calibrating Fake Feature Statistical Distribution for Few-Shot Object Detection (CFSD). First, we utilize feature-specific generative adversarial methods to synthesize fake features, addressing the issue of overfitting in the detection head and classifier when dealing with limited samples. However, due to the model's insufficient training samples, the distribution of the generated fake features in the feature space tends to lean towards the base category. To tackle this issue, we design a distribution calibration module to adjust the statistical distribution of fake features, aligning it with that of novel category features. Finally, we conducted tests on the Pascal VOC and MS COCO datasets, and the results unequivocally confirmed the effectiveness of our method. The test results outperform existing FSOD-based methods and other data augmentation-based methods.

Downloads

Download data is not yet available.

References

[1] Dong X, Zheng L, Ma F, et al. Few-example object detection with model communication [J]. IEEE transactions on pattern analysis and machine intelligence, 2018, 41(7): 1641-1654.

[2] Chen H, Wang Y, Wang G, et al. Lstd: A low-shot transfer detector for object detection [C]//Proceedings of the AAAI conference on artificial intelligence. 2018, 32(1).

[3] Wang Y X, Ramanan D, Hebert M. Meta-learning to detect rare objects [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 9925-9934.

[4] Kang B, Liu Z, Wang X, et al. Few-shot object detection via feature reweighting [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 8420-8429.

[5] Yan X, Chen Z, Xu A, et al. Meta r-cnn: Towards general solver for instance-level low-shot learning [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 9577-9586.

[6] Ni R, Goldblum M, Sharaf A, et al. Data augmentation for meta-learning [C]//International Conference on Machine Learning. PMLR, 2021: 8152-8161.

[7] Rajendran J, Irpan A, Jang E. Meta-learning requires meta-augmentation [J]. Advances in Neural Information Processing Systems, 2020, 33: 5705-5715.

[8] Pei Z, Jiang H, Li X, et al. Data augmentation for rolling bearing fault diagnosis using an enhanced few-shot Wasserstein auto-encoder with meta-learning [J]. Measurement Science and Technology, 2021, 32(8): 084007.

[9] Zhang J, Zhao C, Ni B, et al. Variational few-shot learning [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019: 1685-1694.

[10] Xu J, Le H, Huang M, et al. Variational feature disentangling for fine-grained few-shot classification [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 8812-8821.

[11] Lin X, Duan Y, Dong Q, et al. Deep variational metric learning [C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 689-704.

[12] Gao H, Shou Z, Zareian A, et al. Low-shot learning via covariance-preserving adversarial augmentation networks [J]. Advances in Neural Information Processing Systems, 2018, 31.

[13] Wang Y X, Girshick R, Hebert M, et al. Low-shot learning from imaginary data [C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7278-7286.

[14] Zhang R, Che T, Ghahramani Z, et al. Metagan: An adversarial approach to few-shot learning [J]. Advances in neural information processing systems, 2018, 31.

[15] Li K, Zhang Y, Li K, et al. Adversarial feature hallucination networks for few-shot learning [C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 13470-13479.

[16] Zhao J, Mathieu M, LeCun Y. Energy-based generative adversarial network [J]. arXiv preprint arXiv:1609.03126, 2016.

[17] Tan C, Sun F, Kong T, et al. A survey on deep transfer learning [C]//Artificial Neural Networks and Machine Learning–ICANN 2018: 27th International Conference on Artificial Neural Networks, Rhodes, Greece, October 4-7, 2018, Proceedings, Part III 27. Springer International Publishing, 2018: 270-279.

[18] Zhuang F, Qi Z, Duan K, et al. A comprehensive survey on transfer learning [J]. Proceedings of the IEEE, 2020, 109(1): 43-76.

[19] Qiao L, Zhao Y, Li Z, et al. DeFRCN: Decoupled Faster R-CNN for few-shot object detection [C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021: 8681-8690.

[20] Zhu C, Liang J, Zhou F. Transfer learning-based YOLOv3 model for road dense object detection [J]. Journal of Electronic Imaging, 2023, 32(6): 062505-062505.