Abstract
Though current semantic segmentation methods achieve high accuracy, most of them suffer from low speed, massive memory usage, and high computation complexity. To avoid these problems, we propose a light-weight network called Hybrid Dilated Convolution Network (HDCNet). HDCNet mainly consists of the Hybrid Scale-Aligned Block (HSAB) and the Attentive Depthwise Separable Block (ADSB). The HSAB adopts multiple small kernel convolutions with small-scale dilation rates to extract local information and applies several large kernel convolutions with large-scale dilation rates to encode global information, respectively. We further explore the best option to match the kernel size to the dilation scale. The ADSB is designed to decrease redundant parameters and enhance the critical information by depthwise separable convolution and mixed convolution kernels. In this way, ADSB and HSAB jointly encode multi-scale context information to improve model performance. Thereafter, we combine integrated local information with global information to generate final prediction results. Extensive experiments on Cityscape dataset have demonstrated that the proposed method reaches a better trade-off between accuracy and efficiency compared with other start-of-the-art methods. In particular, HDCNet obtains 72.82% MIoU with only 2.02M and 16.8 GFLOPs.
This work is partially supported by the National Natural Science Foundation of China under Grant No. 61702176.
The first author is a graduate student.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Hong, W.X., Wang, Z.Z., Yang, M., Yuan, J.S.: Conditional generative adversarial network for structured domain adaptation. In: CVPR, pp. 1335–1344 (2018)
Chen, L.C., Papandreou, G., Kokkinos, I., Murphy, K., Yuille, A.L.: DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE TPAMI 40, 834–848 (2018)
Huang, G., Liu, Z., Van Der Maaten, L., Weinberger, K.Q.: Densely connected convolutional networks. In: CVPR, pp. 2261–2269 (2017)
Sandler, M., Howard, A.G., Zhu M.L., Zhmoginov, A., Chen, L.C.: MobileNetV2: inverted residuals and linear bottlenecks. In: CVPR, pp. 4510–4520 (2018)
Mehta, S., Rastegari, M., Shapiro, L.G., Hajishirzi, H.: ESPNetv2: a light-weight, power efficient, and general purpose convolutional neural network. In: CVPR, pp. 9190–9200 (2019)
Li, H.C., Xiong, P.F., Fan, H.Q., Sun, J: Deep feature aggregation for real-time semantic segmentation. In: CVPR, pp. 9522–9531 (2019)
Ding, H.H., Jiang, X.D., Shuai, B., Liu, A.Q., Wang, G.: Context contrasted feature and gated multi-scale aggregation for scene segmentation. In: CVPR, pp. 2393–2402 (2018)
Zhao, H., Shi, J., Qi, X., Wang, X., Jia, J.Y.: Pyramid scene parsing network. In: CVPR, pp. 6230–6239 (2017)
Wang, X.L., Girshick, R.B., Gupta, A., He, K.M.: Non-local neural networks. In: CVPR, pp. 7794–7803 (2018)
Chen, Y.P., Kalantidis, Y., Li, J.S., Yan, S.C., Feng, J.S.: A2-Nets: double attention networks. In: NeurIPS, pp. 350–359 (2018)
Fu, J., Liu, J., Tian, H.J., Li, Y., Bao, Y.J., Fang, Z.W., Lu, H.Q.: Dual attention network for scene segmentation. In: CVPR, pp. 3146–3154 (2019)
Ioffe, S., Szegedy, C.: Batch normalization: accelerating deep network training by reducing internal covariate shift. In: ICML, pp. 448–456 (2015)
He, K.M., Zhang, X.Y., Ren, S.Q., Sun, J.: Delving deep into rectifiers: surpassing human-level performance on imagenet classification. In: ICCV, pp. 1026–1034 (2015)
He, K.M., Zhang, X.Y., Ren, S.Q., Sun, J.: Delving deep into rectifiers: surpassing human-level performance on imagenet classification. In: ICCV, pp. 1026–1034 (2015)
Jiang, B., Tu, W.X., Yang, C., Yuan, J.: Context-integrated and feature-refined network for lightweight object parsing. In: IEEE, pp. 5079–5093 (2020)
Cordts, M., et al.: The cityscapes dataset for semantic urban scene understanding. In: CVPR, pp. 3213–3223 (2016)
Shelhamer, E., Long, J., Darrell, T.: Fully convolutional networks for semantic segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 39(4), 640–651 (2017)
de Oliveira Jr., L.A., et al.: SegNetRes-CRF: a deep convolutional encoder-decoder architecture for semantic image segmentation. In: IJCNN, pp. 1–6 (2018)
Paszke, A., Chaurasia, A., Kim, S., Culurciello, E.: Enet: a deep neural network architecture for real-time semantic segmentation. In: arXiv preprint arXiv:1606.02147 (2016)
Romera, E., Alvarez, J.M., Bergasa, L.M., Arroyo, R.: Erfnet: efficient residual factorized convnet for real-time semantic segmentation. IEEE TITS 19, 263–272 (2018)
Zhao, H.S., Qi, X.J., Shen, X.Y., Shi, J.P., Jia, J.Y.: ICNet for real-time semantic segmentation on high-resolution images. In: arXiv preprint arXiv:1704.08545v2 (2018)
Siam, M., Gamal, M., Abdel-Razek, M., Yogamani, S.K., Jägersand, M.: RTSeg: real-time semantic segmentation comparative study. In: ICIP, pp. 1603–1607 (2018)
Yang, M.K., Yu, K., Zhang, C., Li, Z.W., Yang, K.Y.: DenseASPP for semantic segmentation in street scenes. In: CVPR, pp. 3684–3692 (2018)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
He, J., Jiang, B., Yang, C., Tu, W. (2020). Hybrid Dilated Convolution Network Using Attentive Kernels for Real-Time Semantic Segmentation. In: Peng, Y., et al. Pattern Recognition and Computer Vision. PRCV 2020. Lecture Notes in Computer Science(), vol 12305. Springer, Cham. https://doi.org/10.1007/978-3-030-60633-6_11
Download citation
DOI: https://doi.org/10.1007/978-3-030-60633-6_11
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-60632-9
Online ISBN: 978-3-030-60633-6
eBook Packages: Computer ScienceComputer Science (R0)