End-to-end weakly supervised semantic segmentation with reliable region mining

Zhang, Bingfeng; Xiao, Jimin; Wei, Yunchao; Huang, Kaizhu; Luo, Shan; Zhao, Yao

doi:10.1016/j.patcog.2022.108663

Cited by 29 publications

(5 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Comparison with existing weakly supervised learning framework. We compared our method with two existing weakly supervised learning frameworks: reliable region mining (RRM) [57] and CycleGAN [52]. As shown in Table V, our methods generally produce a higher true positive rate and a lower false positive rate than RRM and cycleGAN, independent of segmentation models (i.e., U-Net, FCN, and DeepLab).…”

Section: Evaluation Of Pseudo Label Generationmentioning

confidence: 99%

“…By generating the initial location seeds, CAM [58] provides a practical solution to solve this issue, and thus has been widely adopted as the first step of weakly supervised learning frameworks. [10], [21], [57]. [57] developed an end-to-end approach named reliable region mining for weakly supervised semantic segmentation.…”

Section: Related Workmentioning

confidence: 99%

“…[10], [21], [57]. [57] developed an end-to-end approach named reliable region mining for weakly supervised semantic segmentation. Combined with CAM, their model applied additional conditional random field operation to get reliable object regions.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Cardiac Adipose Tissue Segmentation via Image-Level Annotations

Huang

Gan

Lye

et al. 2023

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

Automatically identifying the structural substrates underlying cardiac abnormalities can potentially provide real-time guidance for interventional procedures. With the knowledge of cardiac tissue substrates, the treatment of complex arrhythmias such as atrial fibrillation and ventricular tachycardia can be further optimized by detecting arrhythmia substrates to target for treatment (i.e., adipose) and identifying critical structures to avoid. Optical coherence tomography (OCT) is a real-time imaging modality that aids in addressing this need. Existing approaches for cardiac image analysis mainly rely on fully supervised learning techniques, which suffer from the drawback of workload on labor-intensive annotation process of pixelwise labeling. To lessen the need for pixel-wise labeling, we develop a two-stage deep learning framework for cardiac adipose tissue segmentation using image-level annotations on OCT images of human cardiac substrates. In particular, we integrate class activation mapping with superpixel segmentation to solve the sparse tissue seed challenge raised in cardiac tissue segmentation. Our study bridges the gap between the demand on automatic tissue analysis and the lack of high-quality pixel-wise annotations. To the best of our knowledge, this is the first study that attempts to address cardiac tissue segmentation on OCT images via weakly supervised learning techniques. Within an invitro human cardiac OCT dataset, we demonstrate that our weakly supervised approach on image-level annotations achieves comparable performance as fully supervised methods trained on pixel-wise annotations.

show abstract

Section: Evaluation Of Pseudo Label Generationmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Cardiac Adipose Tissue Segmentation via Image-Level Annotations

Huang

Gan

Lye

et al. 2023

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

show abstract

“…Weakly supervised semantic segmentation is a challenging task that only takes imagelevel labels as supervision but produces pixel-level predictions for testing [39]. In this paper, we take orientation labels as the supervision to distinguish the mold and FRC regions.…”

Section: Weakly Supervised Semantic Segmentationmentioning

confidence: 99%

Automated Industrial Composite Fiber Orientation Inspection Using Attention-Based Normalized Deep Hough Network

Zhang

Liang

2023

Micromachines

View full text Add to dashboard Cite

Fiber-reinforced composites (FRC) are widely used in various fields due to their excellent mechanical properties. The mechanical properties of FRC are significantly governed by the orientation of fibers in the composite. Automated visual inspection is the most promising method in measuring fiber orientation, which utilizes image processing algorithms to analyze the texture images of FRC. The deep Hough Transform (DHT) is a powerful image processing method for automated visual inspection, as the “line-like” structures of the fiber texture in FRC can be efficiently detected. However, the DHT still suffers from sensitivity to background anomalies and longline segments anomalies, which leads to degraded performance of fiber orientation measurement. To reduce the sensitivity to background anomalies and longline segments anomalies, we introduce the deep Hough normalization. It normalizes the accumulated votes in the deep Hough space by the length of the corresponding line segment, making it easier for DHT to detect short, true “line-like” structures. To reduce the sensitivity to background anomalies, we design an attention-based deep Hough network (DHN) that integrates attention network and Hough network. The network effectively eliminates background anomalies, identifies important fiber regions, and detects their orientations in FRC images. To better investigate the fiber orientation measurement methods of FRC in real-world scenarios with various types of anomalies, three datasets have been established and our proposed method has been evaluated extensively on them. The experimental results and analysis prove that the proposed methods achieve the competitive performance against the state-of-the-art in F-measure, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE).

show abstract

“…Weakly supervised Pseudo labels Existing methods are broadly classified into two categories. One is a one-stage approach which usually comes to build an end-to-end model and introduces multi-instance learning and some other conditions to optimize it, the overall framework is relatively simple, but the accuracy is generally low as in [13]. The other class of methods is a twostage one, which usually uses some method to generate pseudo-labels first and then uses them for training, such as [14].…”

Section: Related Workmentioning

confidence: 99%

PLDMLT: Multi-Task Learning of Diabetic Retinopathy Using the Pixel-Level Labeled Fundus Images

Liu,

Huang

2023

Computers, Materials &Amp; Continua

View full text Add to dashboard Cite

In the field of medical images, pixel-level labels are time-consuming and expensive to acquire, while image-level labels are relatively easier to obtain. Therefore, it makes sense to learn more information (knowledge) from a small number of hard-to-get pixel-level annotated images to apply to different tasks to maximize their usefulness and save time and training costs. In this paper, using Pixel-Level Labeled Images for Multi-Task Learning (PLDMLT), we focus on grading the severity of fundus images for Diabetic Retinopathy (DR). This is because, for the segmentation task, there is a finely labeled mask, while the severity grading task is without classification labels. To this end, we propose a two-stage multi-label learning weakly supervised algorithm, which generates initial classification pseudo labels in the first stage and visualizes heat maps at all levels of severity using Grad-Cam to further provide medical interpretability for the classification task. A multitask model framework with U-net as the baseline is proposed in the second stage. A label update network is designed to alleviate the gradient balance between the classification and segmentation tasks. Extensive experimental results show that our PLDMLT method significantly outperforms other stateof-the-art methods in DR segmentation on two public datasets, achieving up to 98.897% segmentation accuracy. In addition, our method achieves comparable competitiveness with single-task fully supervised learning in the DR severity grading task.

show abstract

End-to-end weakly supervised semantic segmentation with reliable region mining

Cited by 29 publications

References 11 publications

Cardiac Adipose Tissue Segmentation via Image-Level Annotations

Cardiac Adipose Tissue Segmentation via Image-Level Annotations

Automated Industrial Composite Fiber Orientation Inspection Using Attention-Based Normalized Deep Hough Network

PLDMLT: Multi-Task Learning of Diabetic Retinopathy Using the Pixel-Level Labeled Fundus Images

Contact Info

Product

Resources

About