Advances in Image Segmentation and Uncertainty Quantification

The field of image segmentation is moving towards incorporating spatial correlations and uncertainty quantification to improve model performance and reliability. Conformal prediction is being explored as a means to provide statistically valid uncertainty estimates, which is particularly valuable in high-stakes domains such as medical imaging and aerospace. Recent work has focused on developing methods that can account for spatial structure in image data, resulting in more accurate and interpretable uncertainty estimates. Noteworthy papers include CONSIGN, which proposes a conformal prediction-based method for image segmentation that incorporates spatial correlations, and Robust Vision-Based Runway Detection through Conformal Prediction and Conformal mAP, which applies conformal prediction to quantify localization reliability in runway detection. Additionally, Deep mineralogical segmentation of thin section images based on QEMSCAN maps presents a Convolutional Neural Network model for automatic mineralogical segmentation of thin section images, showing promising results in delineating mineral boundaries and estimating mineral distributions.

Sources

CONSIGN: Conformal Segmentation Informed by Spatial Groupings via Decomposition

From Pixels to Images: Deep Learning Advances in Remote Sensing Image Semantic Segmentation

Robust Vision-Based Runway Detection through Conformal Prediction and Conformal mAP

Deep mineralogical segmentation of thin section images based on QEMSCAN maps

Built with on top of