Coupling Global Context and Local Contents for Weakly-Supervised Semantic Segmentation

Wang, Chunyan; Zhang, Dong; Zhang, Liyan; Tang, Jinhui

doi:10.1109/tnnls.2023.3269513

Cited by 11 publications

(8 citation statements)

References 81 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…According to the characteristics of the skin dataset, the training parameters will also be adjusted accordingly. CONTA [25] uses a causal graph to analyze the relationship between various components in a weakly supervised semantic segmentation model, thereby finding that the essential reason for the inaccuracy of existing pseudo-labels is that the context prior in the dataset is a confounding factor. On this basis, it further proposes to use causal intervention to cut off the connection between context prior and image, thereby improving the quality of pseudo-labels.…”

Section: Methodsmentioning

confidence: 99%

A pixel affinity adaptive expansion weakly supervised skin image segmentation method

huang,

fan

2023

Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023)

View full text Add to dashboard Cite

Image annotation of medical scenes is expensive, then weak supervision methods have emerged. Therefore, this paper proposes a pixel affinity adaptive expansion model for medical image label generation in skin scenes. The method is divided into two stages. In the first stage, initial pseudo label is generated based on the characteristics of the skin image using the proposed self-adaptation expansion module . In the second stage, the generated initial pseudo label uses the characteristics of the pixel pair to distinguish the foreground background and class boundaries to generate the final label. Finally, the generated tags are used for fully supervised segmentation. Experiments were conducted on the ISIC 2016 and ISIC 2017 datasets, respectively. The MIoU of the proposed method was 63.24% and 50.43% on the two datasets, while the Dice-score was 76.13% and 67.79%, the accuracy rate was 81.28% and 64.79%, and the recall rate was 73.54% and 71.03%, which is higher than most existing weakly supervised semantic segmentation algorithms, and achieved Sota (State of the arts) on the two skin datasets.

show abstract

Section: Methodsmentioning

confidence: 99%

A pixel affinity adaptive expansion weakly supervised skin image segmentation method

huang,

fan

2023

Third International Conference on Advanced Algorithms and Signal Image Processing (AASIP 2023)

View full text Add to dashboard Cite

show abstract

“…Li [35] and Wu [56] proposed to extract semantic regions from various input images and discover similar semantic regions. Some recent studies has provided new explanations for CAM generation, including causal inference [61], information bottleneck theory [30], and anti-adversarial attacks [32]. Jiang [25] proposes the unusual use of a straightforward framework (L2G) approach of online local-to-global knowledge transfer.…”

Section: Weakly Supervised Semantic Segmentationmentioning

confidence: 99%

A Weakly Supervised Semantic Segmentation Method based on Local Superpixel Transformation

Chen

et al. 2023

Preprint

View full text Add to dashboard Cite

Weakly supervised semantic segmentation (WSSS) can obtain pseudo-semantic masks through a weaker level of supervised labels, reducing the need for costly pixel-level annotations. However, the general class activation map (CAM)-based pseudo-mask acquisition method suffers from sparse coverage, leading to false positive and false negative regions that reduce accuracy. We propose a WSSS method based on local superpixel transformation that combines superpixel theory and image local information. Our method uses a superpixel local consistency weighted cross-entropy loss to correct erroneous regions and a post-processing method based on the adjacent superpixel affinity matrix (ASAM) to expand false negatives, suppress false positives, and optimize semantic boundaries. Our method achieves 73.4% mIoU on the PASCAL VOC 2012 validation set and 73.9% on the test set, and the ASAM post-processing method is validated on several state-of-the-art methods. If our paper is accepted, our code will be published.

show abstract

“…By causal intervention, the spurious correlations between cause and effect are cut off and the true causality could be accurately estimated (Glymour, Pearl, and Jewell 2016). Many methods taking inspiration from causal inference have been explored to help deep neural networks to learn the true causalities (Zhang et al 2020;Tang, Huang, and Zhang 2020). The true causalities could reduce the prediction error of deep learning models (Niu et al 2021).…”

Section: Related Workmentioning

confidence: 99%

Causal Intervention for Human Trajectory Prediction with Cross Attention Mechanism

Song

Huang

2023

AAAI

View full text Add to dashboard Cite

Human trajectory Prediction (HTP) in complex social environments plays a crucial and fundamental role in artificial intelligence systems. Conventional methods make use of both history behaviors and social interactions to forecast future trajectories. However, we demonstrate that the social environment is a confounder that misleads the model to learn spurious correlations between history and future trajectories. To end this, we first formulate the social environment, history and future trajectory variables into a structural causal model to analyze the causalities among them. Based on causal intervention rather than conventional likelihood, we propose a Social Environment ADjustment (SEAD) method, to remove the confounding effect of the social environment. The core of our method is implemented by a Social Cross Attention (SCA) module, which is universal, simple and effective. Our method has consistent improvements on ETH-UCY datasets with three baseline models and achieves competitive performances with existing methods.

show abstract

Coupling Global Context and Local Contents for Weakly-Supervised Semantic Segmentation

Cited by 11 publications

References 81 publications

A pixel affinity adaptive expansion weakly supervised skin image segmentation method

A pixel affinity adaptive expansion weakly supervised skin image segmentation method

A Weakly Supervised Semantic Segmentation Method based on Local Superpixel Transformation

Causal Intervention for Human Trajectory Prediction with Cross Attention Mechanism

Contact Info

Product

Resources

About