Lightweight Self-Attention Network for Semantic Segmentation

Zhou, Yan; Zhou, Haibin; Li, Nanjun; Li, Jianxun; Wang, Dongli

doi:10.1109/ijcnn55064.2022.9891928

Cited by 3 publications

(4 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The performance of the FCB-SwinV2 Transformer for the Kvasir-SEG [11] dataset is reported in Table 12 for a random data partition. Comparisons to FCN-Transformer model performance from the original paper [7] and from [42] with the advanced data augmentation technique named 'Spatially Exclusive Pasting' (SEP) are provided. Note that Results for the Meta-Polyp model reported in [48] (95.90mDice and 92.10mIoU) are not included in Table 12.…”

Section: Random Data Partitioning Resultsmentioning

confidence: 99%

“…However, for the relatively small image datasets used to evaluate colonoscopy segmentation models, minor differences in how the data is partitioned could cause noticeable performance changes. Recently this has been demonstrated for models being evaluated on the Kvasir-SEG dataset where performance changes greater than 1% were shown to occur for different data partitions [42]. This is significant because the current highest performing models are now typically separated by less than 1% performance differences.…”

Section: A Dataset Selection and Partitioningmentioning

confidence: 95%

See 1 more Smart Citation

Polyp Segmentation With the FCB-SwinV2 Transformer

Fitzgerald,

Bernal,

Histace

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Polyp segmentation within colonoscopy video frames using deep learning models has the potential to automate colonoscopy screening procedures. This could help improve the early lesion detection rate and in vivo characterization of polyps which could develop into colorectal cancer. Recent state-of-theart deep learning polyp segmentation models have combined Convolutional Neural Network (CNN) architectures and Transformer Network (TN) architectures. Motivated by the aim of improving the performance of polyp segmentation models and their robustness to data variations beyond those covered during training, we propose a new CNN-TN hybrid model named the FCB-SwinV2 Transformer. This model was created by making extensive modifications to the recent state-of-the-art FCN-Transformer, including replacing the TN branch architecture with a SwinV2 U-Net. The performance of the FCB-SwinV2Transformer is evaluated on the popular colonoscopy segmentation benchmarking datasets Kvasir-SEG, CVC-ClinicDB and ETIS-LaribPolypDB. Generalizability tests are also conducted to determine if models can maintain accuracy when evaluated on data outside of the training distribution. The FCB-SwinV2 Transformer consistently achieves higher mean Dice and mean IoU scores when compared to other models reported in literature and therefore represents new state-of-the-art performance. The importance of understanding subtleties in evaluation metrics and dataset partitioning are also demonstrated and discussed.

show abstract

Section: Random Data Partitioning Resultsmentioning

confidence: 99%

Section: A Dataset Selection and Partitioningmentioning

confidence: 95%

Polyp Segmentation With the FCB-SwinV2 Transformer

Fitzgerald,

Bernal,

Histace

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…A preliminary conference version [1] of this paper is presented in IJCNN 2022. We extend our previous study with three major improvements.…”

Section: Introductionmentioning

confidence: 99%

“…Fig.3: Architectures of standard self-attention module, disentangled non-local module[39], LSA module[1], and LSA++ module.…”

mentioning

confidence: 99%

LUNet: An enhanced upsampling fusion network with efficient self-attention for semantic segmentation

Zhou,

Yang

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Semantic segmentation is an essential aspect of many computer vision tasks. Self-attention (SA) based deep learning methods have shown impressive results in semantic segmentation by capturing long-range dependencies and contextual information. However, the standard SA module has high computational complexity, which limits its use in resource-constrained scenarios. This paper proposes a novel LUNet to improve semantic segmentation performance while addressing the computational challenges of SA. The lightweight self-attention plus (LSA++) module is introduced as a lightweight and efficient variant of the SA module. LSA++ uses compact feature representation and local position embedding to significantly reduce computational complexity while surpassing the accuracy of the standard SA module. Furthermore, to address the loss of edge details during decoding, we propose the Enhanced Upsampling Fusion Module (EUP-FM). This module comprises an enhanced upsampling module and a semantic vector-guided fusion mechanism. EUP-FM effectively recovers edge information and improves the precision of the segmentation map. Comprehensive experiments on PASCAL VOC 2012, Cityscapes, COCO, and SegPC 2021 demonstrate that LUNet outperforms all compared methods. It achieves superior runtime performance and accurate segmentation with excellent model generalization ability. The code is available at https://github.com/hbzhou530/LUNet.

show abstract

Lunet: an enhanced upsampling fusion network with efficient self-attention for semantic segmentation

Zhou,

Yang

et al. 2024

Vis Comput

View full text Add to dashboard Cite

Lightweight Self-Attention Network for Semantic Segmentation

Cited by 3 publications

References 29 publications

Polyp Segmentation With the FCB-SwinV2 Transformer

Polyp Segmentation With the FCB-SwinV2 Transformer

LUNet: An enhanced upsampling fusion network with efficient self-attention for semantic segmentation

Lunet: an enhanced upsampling fusion network with efficient self-attention for semantic segmentation

Contact Info

Product

Resources

About