U-Net
PulseAugur coverage of U-Net — every cluster mentioning U-Net across labs, papers, and developer communities, ranked by signal.
4 天有情绪数据
-
混合量子-经典模型增强天气降尺度
研究人员开发了一种用于气象降尺度的混合量子-经典扩散模型，将变分量子电路集成到UNet架构中。该方法旨在从粗略输入中增强高分辨率天气数据的重建。初步评估显示，与纯经典模型相比，在平均绝对误差（MAE）和连续排序概率得分（CRPS）方面有所改进，同时保留了大规模空间组织和动能谱。
-
开放式PET/CT基础模型以更少的数据推进肿瘤分割
研究人员开发了一个用于分割FDG PET/CT扫描中肿瘤的开源基础模型，从一开始就整合了解剖学和代谢数据。该模型在近5000个来自多个公共数据集的标准化扫描上进行训练，显示出显著的标签效率，仅用10%的标记数据就能达到与全数据集模型相当的性能。该框架利用了具有早期逐通道连接和掩码自动编码目标的层次化UNet骨干网络，为推进自动化肿瘤成像和减少标注需求提供了坚实的基础。
-
深度学习利用 AlphaEarth 嵌入绘制加州番茄图
研究人员开发了一种新方法，利用 Google DeepMind 的 AlphaEarth 地理空间嵌入和深度学习 U-Net 模型来绘制加利福尼亚的番茄种植系统图。这种方法消除了手动特征工程的需要，而手动特征工程在以前的遥感工作流程中很常见。该模型在独立测试集上实现了高精度，像素精度、精确率、召回率和 F1 分数均超过 99%。模型生成的置信度图在田地边缘附近最高，表明田地内部的预测是可靠的。
-
CryoNet 使用深度学习进行高级冰川测绘
研究人员开发了 CryoNet，一个旨在利用多模态数据测绘碎屑覆盖冰川的深度学习框架。该框架整合了卫星图像、地形数据、光谱指数和雷达信息，以区分裸冰冰川、碎屑覆盖冰川和冰川湖。CryoNet 取得了很高的性能指标，包括 90.52% 的总体 IoU，在复杂山区环境中优于现有的最先进模型。
-
New framework REPA-P enhances physics diffusion models without inference overhead
Researchers have developed a new framework called REPA-P to improve the accuracy and robustness of physics-informed diffusion models. This method aligns intermediate model representations with physical states during tra…
-
Deep Learning Models Achieve High Accuracy in COVID-19 CT Lesion Prediction
Researchers have evaluated deep learning architectures for predicting COVID-19 lesions in CT scans, addressing the lack of standardized performance analysis in medical image segmentation. The study integrated four segme…
-
Deep neural networks enhance urban land surface temperature data resolution
Researchers have developed deep neural networks to improve the resolution of land surface temperature (LST) data for urban areas. By combining data from geostationary and polar-orbiting satellites, they created LST fiel…
-
Neuromorphic depth estimation uses event cameras with uncertainty modeling
Researchers have developed a neuromorphic approach to monocular depth estimation using event cameras, which offer advantages like high temporal resolution and dynamic range. Their deep neural network models predict per-…
-
Cold diffusion tackles percussive audio dereverberation
Researchers have developed a novel cold diffusion framework to address the challenge of dereverberating percussive audio signals, such as drums, which have been largely overlooked in favor of speech processing. This new…
-
TopoU-Net architecture handles complex data structures
Researchers have developed TopoU-Net, a novel U-Net architecture designed to handle complex datasets with higher-order structures beyond simple grids or graphs. This architecture leverages combinatorial complexes, using…
-
New attention module boosts hyperspectral imaging for autonomous driving
Researchers have developed a Multi-Scale Attention Mechanism (MSAM) to improve hyperspectral image segmentation for autonomous driving systems. This module integrates into UNet architectures, using parallel 1D convoluti…
-
AI pipeline accurately segments vocal cord function from video for pathology assessment
Researchers have developed a novel two-stage pipeline for automated glottal area segmentation from high-speed videoendoscopy. This system, which combines a YOLOv8n localizer with a U-Net segmenter, achieved high accurac…
-
Vision transformers outperform CNNs in segmenting cosmic proto-halos
Researchers have developed deep learning models, specifically a U-Net transformer and a V-Net-based CNN, to segment proto-halos in the early universe's density field. The transformer-based network demonstrated superior …
-
Deep Wave Network architecture improves accuracy-cost trade-off for physical dynamics modeling
Researchers have introduced the Deep Wave Network (DW-Net), an architectural innovation for U-Net-type models used in physical dynamics modeling. DW-Net enhances effective depth by stacking multiple encoder-decoder "wav…
-
Pix2Geomodel shows robustness and transferability in complex reservoir modeling
Researchers have developed a Pix2Pix-based model, Pix2Geomodel, to improve reservoir geomodeling by translating between geological facies and petrophysical properties. The model demonstrated robustness and transferabili…
-
Simpler U-Net model outperforms complex attention models for InSAR phase unwrapping
A new paper challenges the trend of using complex computer vision models in InSAR phase unwrapping, demonstrating that a simpler U-Net architecture outperforms attention-based models. The study, conducted on a large InS…
-
New network SANet improves infrared small target detection with attention
Researchers have developed SANet, a novel Selective Attention-based Network designed to improve the detection of small, dim targets in infrared imagery. This network addresses limitations in existing encoder-decoder arc…
-
新AI方法提升时间序列预测的准确性和可解释性
研究人员引入了几种新的时间序列预测方法，旨在提高准确性和泛化能力。MeLISA是一种无潜在变量的自回归模型，可提高回溯效率和长视界统计准确性。Temporal Functional Circuits利用Kolmogorov-Arnold Networks (KANs)为预测提供忠实且与时间相关的解释。Dynamic Pattern Recalibration (DPR)提供了一种与骨干网络无关的令牌级重新校准机制，以适应不断变化的局部…
-
Super-resolution of airborne laser scanning point clouds for forest inventory
Researchers have developed a deep learning model called 3D Forest Super Resolution (3DFSR) to enhance airborne laser scanning (ALS) point clouds for more accurate forest inventory. This voxel-based CNN with a U-Net arch…
-
Researchers develop unsupervised AI for denoising low-dose CT liver scans
Researchers have developed a new unsupervised deep learning framework to denoise low-dose computed tomography (CT) liver scans. This method addresses the challenge of using real clinical data, which is often not suitabl…