DSNet: A Novel Way to Use Atrous Convolutions in Semantic Segmentation
This work addresses the need for faster and more accurate semantic segmentation models for applications like autonomous driving and scene understanding, though it is incremental as it builds on existing atrous convolution methods.
The paper tackles the problem of inefficient use of atrous convolutions in semantic segmentation by proposing DSNet, a dual-branch CNN architecture that incorporates atrous convolutions in shallow layers and uses ImageNet pretraining, achieving state-of-the-art trade-offs in accuracy and speed, such as 40.0% mIOU at 179.2 FPS on ADE20K and 80.4% mIOU at 81.9 FPS on Cityscapes.
Atrous convolutions are employed as a method to increase the receptive field in semantic segmentation tasks. However, in previous works of semantic segmentation, it was rarely employed in the shallow layers of the model. We revisit the design of atrous convolutions in modern convolutional neural networks (CNNs), and demonstrate that the concept of using large kernels to apply atrous convolutions could be a more powerful paradigm. We propose three guidelines to apply atrous convolutions more efficiently. Following these guidelines, we propose DSNet, a Dual-Branch CNN architecture, which incorporates atrous convolutions in the shallow layers of the model architecture, as well as pretraining the nearly entire encoder on ImageNet to achieve better performance. To demonstrate the effectiveness of our approach, our models achieve a new state-of-the-art trade-off between accuracy and speed on ADE20K, Cityscapes and BDD datasets. Specifically, DSNet achieves 40.0% mIOU with inference speed of 179.2 FPS on ADE20K, and 80.4% mIOU with speed of 81.9 FPS on Cityscapes. Source code and models are available at Github: https://github.com/takaniwa/DSNet.