ASFNet: Adaptive multiscale segmentation fusion network for real‐time semantic segmentation

Zha, Hengfeng; Liu, Rui; Yang, Xin; Zhou, Dongsheng; Zhang, Qiang; Wei, Xiaopeng

doi:10.1002/cav.2022

Cited by 9 publications

(7 citation statements)

References 25 publications

(26 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In previous work [8], we found that the actual generalization ability of the adaptive multiscale segmentation fusion module is relatively poor. In contrast, as a very effective structure, residual connection [9] can help the network to back propagate more efficiently and prevent the gradient divergence.…”

Section: Introductionmentioning

confidence: 91%

“…We use Adam as our optimizer with weight decay 2e −4 . For data preprocessing, we used the same approach as ASFNet [8]. For fair comparison with 6…”

Section: Experiments and Analysismentioning

confidence: 99%

“…mIoU: mean Intersection over Union between ground truth and predicted segmentation results. It is calculated based on each category and then the mean value, as shown in Equations (8).…”

Section: Criterionsmentioning

confidence: 99%

See 2 more Smart Citations

Lightweight Real-Time Image Semantic Segmentation Network Based on Multi-Resolution Hybrid Attention Mechanism

Xi-zhong

Liu

Dong

et al. 2022

Wireless Communications and Mobile Computing

Self Cite

View full text Add to dashboard Cite

Effective perception of the surrounding environment and the balance between accuracy and processing speed are crucial for the successful application of real-time semantic segmentation algorithm in the fields of autonomous driving, drones, and smart security. In this paper, a lightweight feature reuse network MHANet for real-time semantic segmentation is proposed. The main novelties of our method are improved ResNet and attention-based fusion mechanism. And the effectiveness of our method is verified by a large number of experiments. Without any pre-training process, the performance of real-time segmentation is improved by using deep fusion of segmentation maps with different resolutions. At the same time, our network converges faster than other networks using pre-training when trained from scratch. Compared with existing methods, the results obtained with our method on the Camvid dataset improve in accuracy (mIoU) ranging from 2% to 6% and in efficiency (FPS) ranging from 15% to 18%. The results achieved 71.87% mIoU of accuracy in the Cityscapes test set, processing images at 203 FPS. Experiments show that manual designed MHANet is effective in improving the performance of real-time semantic segmentation without any pre-training.

show abstract

Section: Introductionmentioning

confidence: 91%

“…We use Adam as our optimizer with weight decay 2e −4 . For data preprocessing, we used the same approach as ASFNet [8]. For fair comparison with 6…”

Section: Experiments and Analysismentioning

confidence: 99%

See 1 more Smart Citation

Lightweight Real-Time Image Semantic Segmentation Network Based on Multi-Resolution Hybrid Attention Mechanism

Xi-zhong

Liu

Dong

et al. 2022

Wireless Communications and Mobile Computing

Self Cite

View full text Add to dashboard Cite

show abstract

“…Multi-level fusion network (MLFNet) 23 performs a feature fusion block to combine spatial details and contextual semantics. Unlike these methods that gradually aggregate features of adjacent stages, adaptive multiscale segmentation fusion network (ASFNet) 27 utilizes an adaptive multiscale module to fuse multi-scale segmentation maps, allowing for effective combination of different levels of features and improved performance.…”

Section: Related Workmentioning

confidence: 99%

PBSNet: pseudo bilateral segmentation network for real-time semantic segmentation

Luo,

Liu,

Mahmoodi

2023

J. Electron. Imag.

View full text Add to dashboard Cite

Achieving real-time performance while maintaining high accuracy in semantic segmentation can be a challenging task. Many existing methods adopt multi-branch architectures to extract both spatial and semantic information, resulting in increased computational complexity and a lack of communication between branches. We propose a pseudo bilateral segmentation network (PBSNet) that can extract rich spatial and semantic features from a single path, without incurring additional computational cost or time consumption. Our proposed PBSNet utilizes a semantic enhancement module to explore the relationship between high-level semantic features, an interchange module to enhance feature representation through bi-directional vertical propagation and adaptive spatial attention, and an attention fusion module to aggregate multi-scale features to produce the final segmentation prediction. Our results on the Cityscapes dataset demonstrate the superiority of PBSNet over state-of-the-art methods, achieving a balance of accuracy and efficiency with 74.52% mean intersection over union and 82.5 frames per second.

show abstract

“…ERFNet [36] innovatively introduced one-dimensional separable residual blocks, replacing each 3×3 convolution with 3×1 and 1×3 convolutions, resulting in a substantial reduction in the number of parameters. ASFNet [37] proposes an adaptive multiscale segmentation fusion network to fuse multiscale contextual to obtain more precise segmentation results. Recent contributions, such as RegSeg [2], have introduced the D-Block, which utilizes two parallel 3×3 convolution layers with different dilation rates to enhance receptive fields.…”

Section: Real-time Semantic Segmentationmentioning

confidence: 99%

BENet: Boundary-Enhanced Network for Real-time Semantic Segmentation

Lei,

Chen,

et al. 2023

Preprint

View full text Add to dashboard Cite

In the realm of real-time semantic segmentation, deep neural networks have demonstrated promising potential. However, current methods face challenges when it comes to accurately segmenting object boundaries and small objects. This limitation is partly attributed to the prevalence of convolutional neural networks, which often involve multiple sequential down-sampling operations, resulting in the loss of fine-grained details. To overcome this drawback, we introduce BENet, a real-time semantic segmentation network with a focus on enhancing object boundaries. The proposed BENet integrates two key components: the Boundary Extraction Module (BEM) and the Boundary Adaption Layer (BAL). The proposed BEM efficiently extracts boundary information, while the BAL guides the network using this information to preserve intricate details during the feature extraction process. Furthermore, to address the challenges associated with poor segmentation of elongated objects, we introduce the Strip Mixed Aggregation Pyramid Pooling Module (SMAPPM). This module employs strip pooling kernels to effectively expand the contextual representation and receptive field of the network, thereby enhancing overall segmentation performance. Our experiments conducted on a single RTX 3090 GPU show that our method achieves an mIoU of 79.4% at a speed of 45.5 FPS on the Cityscapes test set without ImageNet pre-training.

show abstract

ASFNet: Adaptive multiscale segmentation fusion network for real‐time semantic segmentation

Cited by 9 publications

References 25 publications

Lightweight Real-Time Image Semantic Segmentation Network Based on Multi-Resolution Hybrid Attention Mechanism

Lightweight Real-Time Image Semantic Segmentation Network Based on Multi-Resolution Hybrid Attention Mechanism

PBSNet: pseudo bilateral segmentation network for real-time semantic segmentation

BENet: Boundary-Enhanced Network for Real-time Semantic Segmentation

Contact Info

Product

Resources

About