Semantic Segmentation with Labeling Uncertainty and Class Imbalance

  title={Semantic Segmentation with Labeling Uncertainty and Class Imbalance},
  author={Patrik Ol{\~a} Bressan and Jos{\'e} Marcato Junior and Jos{\'e} Augusto Correa Martins and Diogo Nunes Gonçalves and Daniel Matte Freitas and Lucas Prado Osco and Jonathan de Andrade Silva and Zhipeng Luo and Jonathan Li and Raymundo Cordero Garcia and Wesley Nunes Gonçalves},

Figures and Tables from this paper

Hyperbolic Uncertainty Aware Semantic Segmentation

A “Hyperbolic Uncertainty Loss” (HyperUL) is designed, which dynamically highlights the misclassified and high-uncertainty pixels in Hyperbolic space during training via the hyperbolic distances.


Abstract. We present an approach for detecting early signs for upcoming forest damages by training a Convolutional Neural Network (CNN) for the pixel-wise prediction of the remaining life-time (RLT)

Label Adversarial Learning for Skeleton-level to Pixel-level Adjustable Vessel Segmentation

A novel label adversarial learning (LAL) for skeleton-level to pixel-level adjustable vessel segmentation that outperform manual annotations of current public datasets and conventional filtering effects and can be used to generate an uncertainty map representing weak vessel boundaries and noise.

Towards the Identification and Classification of Solar Granulation Structures Using Semantic Segmentation

An appealing potential of the U-net architecture is found to identify cellular patterns in solar granulation images reaching an average accuracy above 80% in the initial training experiments, and this work presents the initial results of a proposed classification model of solar granule structures based on neural semantic segmentation.

Potential applications of deep learning in automatic rock joint trace mapping in a rock mass

In blasted rock slopes and underground openings, rock joints are visible in different forms. Rock joints are often exposed as planes confining rock blocks and visible as traces on a well-blasted,

Transformers for mapping burned areas in Brazilian Pantanal and Amazon with PlanetScope imagery

CFPNet-M: A Light-Weight Encoder-Decoder Based Network for Multimodal Biomedical Image Real-Time Segmentation

Using Convolutional Encoder Networks to Determine the Optimal Magnetic Resonance Image for the Automatic Segmentation of Multiple Sclerosis

Four Convolutional Encoder Networks with different network architectures were used to ensure that the optimal MRI applies to a wide array of deep learning models and determined that FLAIR is the optimal sequence based on Dice Similarity Coefficient (DSC) and Intersection over Union (IoU).



Loss Max-Pooling for Semantic Image Segmentation

A novel loss max-pooling concept for handling imbalanced training data distributions, applicable as alternative loss layer in the context of deep neural networks for semantic image segmentation, and adaptively re-weights the contributions of each pixel based on their observed losses.

Learning to segment under various forms of weak supervision

This work proposes a unified approach that incorporates various forms of weak supervision - image level tags, bounding boxes, and partial labels - to produce a pixel-wise labeling on the challenging Siftflow dataset.

Joint Calibration for Semantic Segmentation

A region-based semantic segmentation framework which handles both full and weak supervision, and addresses three common problems with a joint calibration method which optimizes a multi-class loss defined over the final pixel-level output labeling, as opposed to simply region classification.

Segmentation of Imbalanced Classes in Satellite Imagery using Adaptive Uncertainty Weighted Class Loss

A loss function which leverages the per class uncertainty of the model during training together with median frequency balancing of the class pixels is introduced for the training of deep Convolutional Neural Networks focusing on land use and land cover classification in remote sensed data.

High-performance Semantic Segmentation Using Very Deep Fully Convolutional Networks

A method for high-performance semantic image segmentation based on very deep residual networks, which achieves the state-of-the-art performance and demonstrates that online bootstrapping is critically important for achieving good accuracy.

Label Refinement Network for Coarse-to-Fine Semantic Segmentation

This work proposes a novel network architecture called the label refinement network that predicts segmentation labels in a coarse-to-fine fashion at several resolutions that provides an effective way of producing pixel-wise dense image labeling.

Class-Wise Fully Convolutional Network for Semantic Segmentation of Remote Sensing Images

Experimental results show that the proposed C-FCN significantly improves the segmentation performances compared with many state-of-the-art FCN-based networks, revealing its potentials on accurate segmentation of complex remote sensing images.

Application of Decision Rules for Handling Class Imbalance in Semantic Segmentation

This work adopts a localized method by computing the priors pixel-wise such that the impact can be analyzed at pixel level as well and significantly reduces the non-detection rate for instances of the same classes.

Semantic Segmentation of Tree-Canopy in Urban Environment with Pixel-Wise Deep Learning

The semantic segmentation of trees inside urban environments is highly achievable with deep neural networks, and could be of high importance to decision-making and may contribute to the management of urban systems.

Effective Use of Dilated Convolutions for Segmenting Small Object Instances in Remote Sensing Imagery

A novel architecture called local feature extraction (LFE) module attached on top of dilated front-end module is proposed based on the findings that aggressively increasing dilation factors fails to aggregate local features due to sparsity of the kernel, and detrimental to small objects.