Welcome to Francis Academic Press

Academic Journal of Computing & Information Science, 2025, 8(4); doi: 10.25236/AJCIS.2025.080405.

Lightweight Instance Segmentation Algorithm Based on Knowledge Distillation

Author(s)

Qiangda Shan, Hao Chen, Ziru Liu

Corresponding Author:
Qiangda Shan
Affiliation(s)

School of Information and Artificial Intelligence, Wuhu Institute of Technology, Wuhu, 241000, China

Abstract

Instance segmentation is a technique that performs instance-level and pixel-level segmentation of images simultaneously. this paper proposes an instance-aware segmentation network based on a cross-resolution knowledge distillation architecture. The attention mechanism is utilized to fuse high and low-resolution image features for training the Teacher network, which then guides the Student network - taking low-resolution images as input - in feature extraction, ensuring segmentation accuracy while effectively reducing the network's parameter size and computational load. This paper adopts a pruning method based on switchable normalization to prune the backbone of the network, significantly alleviating the computational pressure during the training of the Teacher network, and further enhancing the real-time performance of the Student network. Experimental results on the public COCO dataset and a self-made pedestrian instance segmentation dataset Person-pic show that the proposed instance segmentation network effectively improves real-time segmentation performance while maintaining segmentation accuracy, achieving a balance between precision and speed.

Keywords

Instance segmentation, knowledge distillation, attention mechanism, network Pruning, lightweight

Cite This Paper

Qiangda Shan, Hao Chen, Ziru Liu. Lightweight Instance Segmentation Algorithm Based on Knowledge Distillation. Academic Journal of Computing & Information Science (2025), Vol. 8, Issue 4: 40-48. https://doi.org/10.25236/AJCIS.2025.080405.

References

[1] Garcia-Garcia, Alberto, Orts-Escolano, Sergio, Oprea, Sergiu, et al. A survey on deep learning techniques for image and video semantic segmentation. Applied Soft Computing, 70: 41--65, 2018.

[2] Woo, Sanghyun, Park, Jongchan, Lee, Joon-Young, et al. Cbam: Convolutional block attention module. //Proceedings of the European conference on computer vision (ECCV), 2018.

[3] Luo, Ping, Zhang, Ruimao, Ren, Jiamin, et al. Switchable Normalization for Learning-to-Normalize Deep Representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(2): 712-728, 2021.

[4] Krizhevsky, Alex, Sutskever, Ilya, Hinton, Geoffrey E. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 2012.

[5] Bharath Hariharan, Pablo Arbeláez, Ross Girshick, Jitendra Malik. Simultaneous detection and segmentation. //European conference on computer vision, 2014. 

[6] Arbel, Pablo, Pont-Tuset, Jordi, Barron, Jonathan T, et al. Multiscale combinatorial grouping. //Proceedings of the IEEE conference on computer vision and pattern recognition, 2014.

[7] He, Kaiming, Gkioxari, Georgia, Dollar, Piotr, et al. Mask r-cnn. //Proceedings of the IEEE international conference on computer vision, 2017.

[8] Bolya, Daniel, Zhou, Chong, Xiao, Fanyi, et al. Yolact: Real-time instance segmentation. //Proceedings of the IEEE/CVF international conference on computer vision, 2019.

[9] Bolya, Daniel, Zhou, Chong, Xiao, Fanyi, et al. Yolact++: Better real-time instance segmentation. IEEE transactions on pattern analysis and machine intelligence, 2020.

[10] Wang, Xinlong, Kong, Tao, Shen, Chunhua, et al. Solo: Segmenting objects by locations. //European Conference on Computer Vision, 2020.

[11] Wang, Xinlong, Zhang, Rufeng, Kong, Tao, et al. Solov2: Dynamic and fast instance segmentation. Advances in Neural information processing systems, 33: 17721--17732, 2020.

[12] Geoffrey, Vinyals, Oriol, Dean, Jeff. Distilling the Knowledge in a Neural Network. Computer Science, 14(7): 38-39, 2015.

[13] Qi, Lu, Kuen, Jason, Gu, Jiuxiang, et al. Multi-scale aligned distillation for low-resolution detection. //Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021.

[14] Tian, Zhi, Shen, Chunhua, Chen, Hao. Conditional convolutions for instance segmentation. //European Conference on Computer Vision, 2020.

[15] He, Kaiming, Zhang, Xiangyu, Ren, Shaoqing, et al. Deep residual learning for image recognition. //Proceedings of the IEEE conference on computer vision and pattern recognition, 2016.

[16] Jiahui Yu, Linjie Yang, Ning Xu, et al. Slimmable Neural Networks. CoRR, abs/1812.08928, 2018.

[17] Jie Hu, Li Shen, Gang Sun. Squeeze-and-Excitation Networks. CoRR, abs/1709.01507, 2017.

[18] Milletari, Fausto, Navab, Nassir, Ahmadi, Seyed-Ahmad. V-net: Fully convolutional neural networks for volumetric medical image segmentation. //2016 fourth international conference on 3D vision (3DV), 2016.

[19] Tian, Zhi, Shen, Chunhua, Wang, Xinlong, et al. Boxinst: High-performance instance segmentation with box annotations. //Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021.

[20] Zhang, Rufeng, Tian, Zhi, Shen, Chunhua, et al. Mask encoding for single shot instance segmentation. //Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.

[21] Tian, Zhi, Shen, Chunhua, Chen, Hao. Conditional convolutions for instance segmentation. //European Conference on Computer Vision, 2020.

[22] Chen, Hao, Sun, Kunyang, Tian, Zhi, et al. Blendmask: Top-down meets bottom-up for instance segmentation. //Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020.