Variance-Entropy Guided Fine-Grained Prototypes Distillation for Robust Few-Shot Object Detection

Adib, Mohammad Hussein; Fadaeieslam, Mohammad Javad

doi:10.22075/mseee.2026.41071.1270

Variance-Entropy Guided Fine-Grained Prototypes Distillation for Robust Few-Shot Object Detection

Articles in Press

Document Type : Research Article

Authors

Mohammad Hussein Adib

Mohammad javad Fadaeieslam

Faculty of Electrical & Computer Engineering, Semnan University, Semnan, Iran.

10.22075/mseee.2026.41071.1270

Abstract

Few-shot object detection (FSOD) aims to enable models to recognize novel objects from minimal training examples; however, the task faces significant hurdles due to unstable feature learning and inherently biased class prototypes when training data is severely limited. Standard detectors often struggle to generalize in these scenarios, as they frequently overfit to the limited support set, leading to poor intra-class compactness. To mitigate these issues, we propose a novel Fine-Grained Prototype Distillation (FPD) framework that utilizes a unique variance-entropy guided refinement strategy. Our approach introduces two core components: first, a variance-aware module for Prototype Distillation (PD) that enhances prototype discriminability through adaptive intra-class dispersion weighting; and second, an entropy-based mechanism for Prototype Assignment (PA) that effectively suppresses low-confidence support samples during training. By explicitly accounting for the statistical reliability of support features, FPD fosters more stable and representative prototype construction, effectively bridging the critical gap between noisy, limited support instances and robust, reliable class representations. Extensive experimental results on the PASCAL VOC benchmark demonstrate that our method consistently achieves superior average precision (AP) compared to state-of-the-art FSOD models in both 1-shot and 5-shot scenarios. These results confirm the effectiveness of our framework, providing a more robust foundation for accurate category recognition in extreme, data-scarce regimes.

Keywords

Classification Confidence

Prototype Refinement

Similarity Weighting

Subjects

Computer

[1]	K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," in Computer Vision and Pattern Recognition, Boston, 2015.
[2]	Z. Wang, B. Yang, H. Yue, and Z. Ma, "Fine-Grained Prototypes Distillation for Few-shot Object Detection," in AAAI, Vancouver, 2024.
[3]	Q. Fan and W. Zhuo, "Few-Shot Object Detection via Variational Feature Aggregation," in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, Waikoloa, 2022.
[4]	K. Chen, "MMDetection: Open MMLab Detection Toolbox and Benchmark," in arXiv, Washington, 2020.
[5]	O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, and A. Karpathy, "ImageNet Large Scale Visual Recognition Challenge," IJCV, vol. 115, no. 3, pp. 211-252, 2015.
[6]	T. Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature Pyramid Networks for Object Detection," in Computer Vision and Pattern Recognition (CVPR), Honolulu, 2017.
[7]	X. Fan, H. Yang, and J. Yan, "Meta-RCNN: Meta Learning for Few-Shot Object Detection," in Proceedings of the 28th ACM International Conference on Multimedia, Washington, 2020.
[8]	Y. Xiao, V. Lepetit, and R. Marlet, "Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild," in Computer Vision and Pattern Recognition, Seattle, 2020.
[9]	X. Hu, "Dense Relation Distillation with Context-aware Aggregation for Few-Shot Object Detection," in CVPR, Virtual, 2021.
[10]	A. Amjoud and M. Amrouch, "Object Detection Using Deep Learning, CNNs and Vision Transformers: A Review," IEEE, pp. 35479-35516, 2023.
[11]	Z. Zou, K. Chen, Z. Shi, Y. Guo, and J. Ye, "Object Detection in 20 Years: A Survey," Proceedings of the IEEE, pp. 257-276, 2023.
[12]	R. Q. Qu and H. Liang, "A Survey of Deep Learning-Based Object Detection," IEEE, p. 128837–128868, 2019.
[13]	B. Kang and Y. Xue, "Few-Shot Object Detection: A Survey," ACM, pp. 1-41, 2022.
[14]	Y. Song and T. Y. Wang, "A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities," ACM, pp. 1-57, 2023.
[15]	X. Zhang, X. Tan, and X. Weng, "Few-Shot Object Detection: Research Advances and Challenges," IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 4608-4627, 2022.
[16]	C. Sun and Y. Li, "Dual-Attention Model for Camera-Based Few-Shot Object Detection," IEEE Sensors, pp. 2045-2059, 2025.
[17]	T. Wang, "A Generalized Few-Shot Object Detection Method via Extraction of Base-Novel Commonality With Memory Distillation of Category Prototypes," IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 7, pp. 456-467, 2025.
[18]	R. Xiao and L. Lepetit, "Few-Shot Object Detection and Viewpoint Estimation for Objects in Wild," IEEE, vol. 23, no. 4, pp. 2045-2052, 2022.
[19]	J. He, "Recent Few-shot Object Detection Algorithms: A Survey with Performance Comparison," Springer, vol. 54, no. 13, pp. 9043-9064, 2024.
[20]	S. Zhang and W. Wang, "Extreme R-CNN: Few-Shot Object Detection via Sample Synthesis and Knowledge Distillation," Sensors, vol. 24, no. 23, pp. 9023-9048, 2024.
[21]	N. Carion, F. Massa, and G. Synnaeve, "End-to-End Object Detection with Transformers," in Computer Vision and Pattern Recognition, 2020.