DCS-TransUperNet: Road Segmentation Network Based on CSwin Transformer with Dual Resolution

Zhang, Zheng; Miao, Chunle; Liu, Changan; Tian, Qing

doi:10.3390/app12073511

Cited by 30 publications

(14 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is due to the characteristics of dilated convolutions, where adjacent pixels are convolved from independent subsets, lacking interdependence. When the dilation rates are set to (1,2,3,4,5,6), although the prediction results have high accuracy, the overall model size increases significantly, contradicting our lightweight design principles.…”

Section: Grid Effectsmentioning

confidence: 83%

Mix-net: A Dual-Branch Lightweight Semantic Segmentation Network Based on MRF

Tan

Zhao

Cai

et al. 2023

Preprint

View full text Add to dashboard Cite

To enable image segmentation tasks to go beyond microcomputers and similar devices and better adapt to the limited computing and storage capabilities of mobile devices, we have proposed a lightweight semantic segmentation network called Mix-net. Firstly, we designed a Mix Convolution (MC) module for feature extraction and recovery, which significantly reduces the number of parameters in the model while maintaining accuracy. Secondly, to enhance the edge details of various semantics in the images, we simultaneously train two network models. One model focuses on learning the main body details of the images, while the other model is specifically designed for segmentation tasks and is more dedicated to capturing the edges of different semantics. Thirdly, we designed a novel activation function called Mix Rectified Function (MRF), which effectively enhances the accuracy of segmentation tasks and is more suitable for lightweight image segmentation tasks. Additionally, we have employed Group Normalization (GN) to complement MRF for normalization and activation operations, enabling higher accuracy for small-batch segmentation tasks. Finally, our proposed network retains skip-connection structures and incorporates channel shuffling and attention modules to enhance accuracy. We have conducted comparative experiments on the Cityscapes dataset, where Mix-net achieved an Mean Intersection over Union (MIoU) of 81.1% and a smaller model size, utilizing only 1.47M Parameters and 9.93G Floating Point Operations (FLOPs) on a single NVIDIA GeForce RTX3060 graphics card.

show abstract

Section: Grid Effectsmentioning

confidence: 83%

Mix-net: A Dual-Branch Lightweight Semantic Segmentation Network Based on MRF

Tan

Zhao

Cai

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…To verify the performance of HA-RoadFormer in road segmentation task, we select representative road segmentation networks, such as DeepLabV3+ [50], D-LinkNet [10], D-ResUnet [51], and Transformer based networks DCS-TransUperNet [26] and ViT [11], making comparative experiments on Massachusetts road dataset.…”

Section: Experimental Results On Massachusetts Road Datasetmentioning

confidence: 99%

“…Recently, the road segmentation framework based on Transformer has also been innovated. For example, Zhang et al [26] propose a remote sensing road segmentation framework based on CSWin Transformer, improving IoU. However, the inference speed is relatively slow compared with the model based on CNNs.…”

Section: Road Segmentation In Remote Sensing Imagementioning

confidence: 99%

“…Before training, we use data enhancement, such as random rotation, to expand the data again. The sum of cross entropy loss and dice loss function is used in the training process as the final loss function [26] and optimized with Adam [45] algorithm. We used 4 NVIDIA RTX 2080ti GPUs (11 G) by a workstation with four graphic card slots.…”

Section: Implementation Detailsmentioning

confidence: 99%

See 1 more Smart Citation

HA-RoadFormer: Hybrid Attention Transformer with Multi-Branch for Large-Scale High-Resolution Dense Road Segmentation

et al. 2022

Self Cite

View full text Add to dashboard Cite

Road segmentation is one of the essential tasks in remote sensing. Large-scale high-resolution remote sensing images originally have larger pixel sizes than natural images, while the existing models based on Transformer have the high computational cost of square complexity, leading to more extended model training and inference time. Inspired by the long text Transformer model, this paper proposes a novel hybrid attention mechanism to improve the inference speed of the model. By calculating several diagonals and random blocks of the attention matrix, hybrid attention achieves linear time complexity in the token sequence. Using the superposition of adjacent and random attention, hybrid attention introduces the inductive bias similar to convolutional neural networks (CNNs) and retains the ability to acquire long-distance dependence. In addition, the dense road segmentation result of remote sensing image still has the problem of insufficient continuity. However, multiscale feature representation is an effective means in the network based on CNNs. Inspired by this, we propose a multi-scale patch embedding module, which divides images by patches with different scales to obtain coarse-to-fine feature representations. Experiments on the Massachusetts dataset show that the proposed HA-RoadFormer could effectively preserve the integrity of the road segmentation results, achieving a higher Intersection over Union (IoU) 67.36% of road segmentation compared to other state-of-the-art (SOTA) methods. At the same time, the inference speed has also been greatly improved compared with other Transformer based models.

show abstract

“…Each pixel in the image can be treated as a position encoding via transformers, and the self-attention mechanism can be utilized to learn pixel relationships, thus better capturing the road features. 18,19 Zhang et al 20 introduced DCS-TransUperNet, which combines transformer with a feature fusion module, to effectively address road segmentation challenges. Hu et al 21 proposed a multi-scale deformable transformer and achieved more precise road segmentation in remote sensing images.…”

Section: Introductionmentioning

confidence: 99%

DELFormer: detail-enhanced lightweight transformer for road segmentation

Xin,

Fu,

et al. 2023

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

.The road segmentation task has become increasingly important in fields such as urban planning, traffic management, and environmental monitoring. However, most existing deep learning-based methods suffer from issues such as poor temporal effectiveness and connectivity, making it a significant challenge to achieve high-precision and high-efficiency road segmentation. We propose a road segmentation model based on a detail-enhanced lightweight transformer. Through the connectivity enhancement module, the issue of spatial information loss is addressed, enhancing the modeling capability of the road network connectivity. The model incorporates a detail-enhancement strategy to capture the relationship between roads and the environment, enhancing the perception and expression of details while maintaining low computational complexity. Furthermore, the use of a lightweight multiple feature fusion module promotes information fusion from features at different scales while a maintaining lightweight design. Extensive experiments on two publicly available datasets demonstrate that our method achieves the best performance in terms of real-time effectiveness and accuracy.

show abstract

DCS-TransUperNet: Road Segmentation Network Based on CSwin Transformer with Dual Resolution

Cited by 30 publications

References 37 publications

Mix-net: A Dual-Branch Lightweight Semantic Segmentation Network Based on MRF

Mix-net: A Dual-Branch Lightweight Semantic Segmentation Network Based on MRF

HA-RoadFormer: Hybrid Attention Transformer with Multi-Branch for Large-Scale High-Resolution Dense Road Segmentation

DELFormer: detail-enhanced lightweight transformer for road segmentation

Contact Info

Product

Resources

About