Mixed-UNet: Refined Class Activation Mapping for Weakly-Supervised Semantic Segmentation with Multi-scale Inference

Liu, Yang; Zhang, Ersi; Xu, Lulu; Xiao, Chufan; Zhong, Xiaoyun; Lijin, Lian,; Li, Fang; Jiang, Bin; Dong, Yuhan; Ma, Li; Huang, Qiming; Xu, Min; Zhang, Yongbing; Yu, Dongmei; Yan, Chenggang; Qin, Peiwu

doi:10.48550/arxiv.2205.04227

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In comparison to methods relying on stronger supervisory information, such as points, annotations, borders, and pixel-level labels, our approach significantly narrows the performance gap. Deeplab [21] Pixel-level VOC 67.3 70.1 What'sPoint [22] Point-level VOC 46.3 43.9 ScribbleSup [23] Graffiti VOC 63.0 一 SDI [24] Boundary VOC 65.7 67.4 CAMs [25] Class labels VOC 71.2 70.3 Mixed-UNet [26] Class labels VOC 70.1 71.3 SAN [27] Class labels VOC 69.9 71.1 MTL [28] Class labels VOC 70.5 70.1 DCNNs [29] Class labels VOC 72.4 73.1 TSEG [30] Class labels VOC 68. The favorable experimental performance of this method can be attributed to the polarized attention mechanism proposed in this paper.…”

Section: Dvoc2012 Datasetmentioning

confidence: 99%

“…Validation of our proposed method on the COCO dataset, under the same hardware conditions and software configurations, is presented in Table V. Mixed-UNet [26] Class labels COCO 33.2 SAN [27] Class labels COCO 36.4 MTL [28] Class labels COCO 44.4 DCNNs [29] Class labels COCO 35.2 TSEG [30] Class labels COCO 40.6 Ours Class labels COCO 45.9…”

Section: Ems-coco Datasetmentioning

confidence: 99%

See 1 more Smart Citation

Polarized Attention Weak Supervised Semantic Segmentation Network

Dai,

Wu,

Dawei

2024

IEEE Access

View full text Add to dashboard Cite

Currently, weakly supervised semantic segmentation methods based on image-level annotation often rely on pseudo pixel masks generated by seed regions. However, the growth of seed regions is stochastic, and in cases where targets are occluded or overlapped in the image without additional reference information, the segmentation network may encounter issues of missed or incorrect segmentation. To address this problem, this paper proposes a polarized attention mechanism for weakly supervised semantic segmentation networks. The attention mechanism consists of a semantic perception branch and a boundary detection branch. The semantic perception branch allows the network to better distinguish the category of each pixel in the image. Subsequently, the boundary detection branch enables the seed region to naturally expand towards the target boundary. The pseudo pixel mask generated by this method provides better coverage of the target area and improves the performance of the segmentation network. The test set and validation set mean Intersection over Union (mIoU) of the PASCAL VOC 2012 dataset achieved 72.9% and 73.2%. The results of the experiments demonstrated the effectiveness of the proposed method. The experimental results indicated that the attention mechanism, as proposed in this paper, can effectively enhance the segmentation performance in situations where objects in the image are occluded or overlapped.

show abstract

Section: Dvoc2012 Datasetmentioning

confidence: 99%

Section: Ems-coco Datasetmentioning

confidence: 99%

Polarized Attention Weak Supervised Semantic Segmentation Network

Dai,

Wu,

Dawei

2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Saab et al (2019) and Salehinejad et al (2021) utilized ResNet-like architectures for binary ICH detection with attention layers and Grad-CAM techniques, respectively, but they only visualized attention and class activation maps for qualitative assessment of their methods. Furthermore, very limited attempts were also made to apply the attention/class activation in weakly supervised brain lesion and hemorrhage segmentation (Wu et al, 2019;Nemcek et al, 2021;Liu et al, 2022b). Specifically, Wu et al (2019) used refined 3D CAMs to segment stroke lesions from the Ischemic Stroke Lesion Segmentation (ISLES) dataset (multi-spectral MRI), and achieved a 0.3827 mean Dice score.…”

Section: Related Workmentioning

confidence: 99%

Weakly Supervised Intracranial Hemorrhage Segmentation using Head-Wise Gradient-Infused Self-Attention Maps from a Swin Transformer in Categorical Learning

Rasoulian,

Salari,

Xiao

2023

Melba

View full text Add to dashboard Cite

Intracranial hemorrhage (ICH) is a life-threatening medical emergency that requires timely and accurate diagnosis for effective treatment and improved patient survival rates. While deep learning techniques have emerged as the leading approach for medical image analysis and processing, the most commonly employed supervised learning often requires large, high-quality annotated datasets that can be costly to obtain, particularly for pixel/voxel-wise image segmentation. To address this challenge and facilitate ICH treatment decisions, we introduce a novel weakly supervised method for ICH segmentation, utilizing a Swin transformer trained on an ICH classification task with categorical labels. Our approach leverages a hierarchical combination of head-wise gradient-infused self-attention maps to generate accurate image segmentation. Additionally, we conducted an exploratory study on different learning strategies and showed that binary ICH classification has a more positive impact on self-attention maps compared to full ICH subtyping. With a mean Dice score of 0.44, our technique achieved similar ICH segmentation performance as the popular U-Net and Swin-UNETR models with full supervision and outperformed a similar weakly supervised approach using GradCAM, demonstrating the excellent potential of the proposed framework in challenging medical image segmentation tasks. Our code is available at <a href='https://github.com/HealthX-Lab/HGI-SAM'>https://github.com/HealthX-Lab/HGI-SAM</a>

show abstract

Mixed-UNet: Refined Class Activation Mapping for Weakly-Supervised Semantic Segmentation with Multi-scale Inference

Cited by 2 publications

References 0 publications

Polarized Attention Weak Supervised Semantic Segmentation Network

Polarized Attention Weak Supervised Semantic Segmentation Network

Weakly Supervised Intracranial Hemorrhage Segmentation using Head-Wise Gradient-Infused Self-Attention Maps from a Swin Transformer in Categorical Learning

Contact Info

Product

Resources

About