Fully convolutional network with attention modules for semantic segmentation

Huang, Yunjia; Xu, Haixia

doi:10.1007/s11760-020-01828-8

Cited by 15 publications

(3 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, we also achieve 69.5% mIoU on test set with 112 FPS. Compared with other models, including PSPNet [16] (412.2 GFOLPs and 250.8 M parameters), DeepLab [21] (457.8 GPOLPs and 262.1 M parameters), and FCN+PPAM+SAM [36] (38.7 GPOLPs and 42.41 M parameters), our model is considerably small.…”

Section: Qualitative Analysis Of Segmentation Resultsmentioning

confidence: 86%

A Hierarchical Feature Extraction Network for Fast Scene Segmentation

Liu

Zhang

2021

Sensors

View full text Add to dashboard Cite

Semantic segmentation is one of the most active research topics in computer vision with the goal to assign dense semantic labels for all pixels in a given image. In this paper, we introduce HFEN (Hierarchical Feature Extraction Network), a lightweight network to reach a balance between inference speed and segmentation accuracy. Our architecture is based on an encoder-decoder framework. The input images are down-sampled through an efficient encoder to extract multi-layer features. Then the extracted features are fused via a decoder, where the global contextual information and spatial information are aggregated for final segmentations with real-time performance. Extensive experiments have been conducted on two standard benchmarks, Cityscapes and Camvid, where our network achieved superior performance on NVIDIA 2080Ti.

show abstract

Section: Qualitative Analysis Of Segmentation Resultsmentioning

confidence: 86%

A Hierarchical Feature Extraction Network for Fast Scene Segmentation

Liu

Zhang

2021

Sensors

View full text Add to dashboard Cite

show abstract

“…In recent years, RCNU-Net, proposed by Boxiong Huang, Tongyuan Huang, et al precisely leverages this concept. [26] Its primary architecture is built upon the traditional U-Net framework, wherein the inclusion of CBAM in skip connections addresses the semantic gaps within multiscale polyp features, thereby enriching the content of RCNU-Net for polyp segmentation.…”

Section: Image Segmentation Networkmentioning

confidence: 99%

PASK-Net: A Hybrid Network for Polyps Image Segmentation

Lv,

Lu,

2024

Preprint

View full text Add to dashboard Cite

Accurate segmentation of polyps is crucial in the field of medical image recognition. Attention mechanisms have been widely applied in medical image segmentation, but attention mechanisms implemented by convolution are limited in capturing multi-scale information due to the constraints of convolution kernel sizes. This linear aggregation method restricts the network's adaptability to various complex situations and has limitations in handling multi-scale information. Therefore, this paper proposes PASK-Net, which performs serial computations in both channel and spatial dimensions. In the channel dimension, a nonlinear approach is introduced to achieve neuron-adaptive receptive field sizes, addressing the problem of fixed convolutional processing range in attention mechanisms that hinders effective acquisition of multi-scale information. Meanwhile, in the spatial dimension, Channel-Prioritized Convolutional Attention (CPCA) is introduced to enhance the network's feature representation capability by aggregating multi-scale information from different-sized convolutional kernel branches, while ensuring computational efficiency. Comparative experiments on the Kvasir dataset demonstrate that, compared to other networks, the PASK module performs well on six evaluation metrics including Dice and IOU, achieving a Dice value of 87.54% and an IOU value of 80.6%, effectively improving the accuracy of polyp segmentation. Results from ablation experiments also validate the effectiveness of the module. The codes are available at https://github.com/LvYamKun/PASK-Net

show abstract

“…Recent transfer learning methods [11][12][13], in which the main adaptation method [5,14] improves generalization of unlabeled target data by aligning distribution. It has been applied in various applications, such as image classification [32], semantic segmentation [25,33], and object detection [4,5]. However, it cannot learn discriminant features in the target domain.…”

Section: Related Workmentioning

confidence: 99%

Zero-Augmentation Pseudo Entropy for Video Unsupervised Domain Adaptation

Chen

Fan

et al. 2022

Preprint

View full text Add to dashboard Cite

This work focuses on the problem of video unsupervised domain adaptation (VUDA) for the task of human action recognition. Although there exist some methods for VUDA, however, most current methods exhibit ambient light bias. Transferring models from light environment to the dark will lead to a decline in performance due to inevitable domain shift. Unsupervised domain adaptation (UDA) methods are able to alleviate the performance decline by aligning source and target domains utilizing methods such as adversarial training. However, these methods do not enhance images for the dark environment, leading to the loss of information and weak contribution to model performance in the dark environment. In this work, we enhance images by specific approach as an auxiliary task for VUDA in addition to unsupervised adver-sarial alignment as our proposed method, Zero-Augmentation Pseudo Entropy (Z-APE). We test our method on the HMDB-ARID datasets and empirically show that our VUDA method improves the performance by average 6.3% compared to other typical VUDA baseline models.

show abstract

Fully convolutional network with attention modules for semantic segmentation

Cited by 15 publications

References 19 publications

A Hierarchical Feature Extraction Network for Fast Scene Segmentation

A Hierarchical Feature Extraction Network for Fast Scene Segmentation

PASK-Net: A Hybrid Network for Polyps Image Segmentation

Zero-Augmentation Pseudo Entropy for Video Unsupervised Domain Adaptation

Contact Info

Product

Resources

About