Cervical cancer diagnosis hinges significantly on precise nuclei segmentation at early stages, which however, remains largely elusive due to challenges such as overlapping cells and blurred nuclei boundaries. This paper presents a novel deep neural network (DNN), the Global Context UNet (GC-UNet), designed to adeptly handle intricate environments and deliver accurate cell segmentation.
1
1 February 2025 in Biomedical Signal Processing and Control
Advancements in computer vision and deep learning have significantly propelled progress in scene understanding, aiding rescue teams in accurately assessing damage after natural disasters. In this paper, we introduce EarthquakeNet, a meticulously curated high-resolution post-earthquake dataset featuring detailed classification and semantic segmentation annotations, designed to enhance comprehensive scene understanding following natural disasters. EarthquakeNet comprises post-disaster images captured using unmanned aerial vehicles (UAVs) from multiple affected areas after an earthquake. The uniqueness of EarthquakeNet lies in providing high-resolution post-disaster imagery, each with exhaustive annotations. Unlike existing datasets that offer annotations for specific scene elements like buildings, EarthquakeNet provides pixel-level annotations for a broader range of categories, including roads, houses, and tents. We also demonstrate the utility of the dataset by implementing state-of-the-art segmentation models on EarthquakeNet, showcasing its value in improving existing methods for natural disaster damage assessment.计算机视觉和深度学习领域的进步极大地推动了场景理解的发展,帮助救援队在自然灾害后准确评估损失。在这篇论文中,我们介绍了EarthquakeNet,这是一个精心挑选的高分辨率地震后数据集,具有详细的分类和语义分割注释,旨在增强自然灾害后的全面场景理解。EarthquakeNet包含地震后使用无人机在多个受灾地区拍摄的图像。EarthquakeNet的独特之处在于提供了高分辨率的地震后影像,每张影像都有详尽的注释。与提供特定场景元素(如建筑物)注释的现有数据集不同,EarthquakeNet为更广泛的类别(包括道路、房屋和帐篷)提供了像素级注释。我们还通过在EarthquakeNet上实施最先进的分割模型,展示了该数据集的实用性,展示了其在改进现有自然灾害损失评估方法方面的价值。