Swin Transformer
PulseAugur coverage of Swin Transformer — every cluster mentioning Swin Transformer across labs, papers, and developer communities, ranked by signal.
1 天有情绪数据
-
SMIT method leads in transferability for medical image segmentation
Researchers have benchmarked nine self-supervised learning (SSL) methods for their transferability in medical image segmentation tasks. The study found that the Self-Distilled Masked Image Transformer (SMIT) method, whi…
-
AI model accurately detects rectal tumor regrowth from endoscopy images
Researchers have developed a novel Siamese Swin Transformer with Dual Cross-Attention (SSDCA) designed to detect local regrowth of rectal tumors from endoscopic images. This model analyzes sequential images from patient…
-
New MorphoFormer AI model improves building height and footprint estimation
Researchers have developed MorphoFormer, a novel framework for jointly estimating building height and footprint using remote sensing data. This approach explicitly encodes the relationship between these two parameters, …
-
TwistNet-2D learns second-order channel interactions for texture recognition
Researchers have developed TwistNet-2D, a novel module designed to enhance texture recognition by capturing second-order channel interactions. This module computes local pairwise channel products with directional spatia…
-
InfiltrNet结合CNN和Transformer用于脑肿瘤浸润风险预测
研究人员开发了InfiltrNet，一种用于预测脑肿瘤浸润风险的新型双分支架构。该系统结合了CNN编码器和Swin Transformer编码器，利用交叉注意力融合从多模态MRI扫描生成风险图。该方法旨在通过估算可见肿瘤边界以外的浸润情况来改进手术规划和放射治疗，在BraTS 2020和BraTS 2025数据集的实验中表现优于现有方法。
-
视觉Transformer利用DCT提升注意力和效率
研究人员开发了一种利用离散余弦变换（DCT）来增强视觉Transformer的新颖方法。该方法包括一种基于DCT的自注意力初始化策略，可提高在CIFAR-10和ImageNet-1K等基准测试上的分类准确性。此外，一种基于DCT的注意力压缩技术通过截断输入块的高频分量来降低计算开销，从而在Swin Transformer等模型中保持性能。
-
New MSR framework improves CT-MRI cervical spine registration with hybrid modeling
Researchers have developed a new framework called MSR for rigid-deformable hybrid modeling in CT-MRI registration of the cervical spine. This approach combines rigid alignment of individual vertebrae with deformable mod…
-
100,000 Yuan Investment: Latest Interview with Princeton's Zhuang Liu: Architecture Isn't That Important, Data is King
Princeton Assistant Professor Liu Zhuang argues that AI architecture is less critical than previously thought, with data scale and diversity being the primary drivers of progress. In a recent interview, he highlighted t…
-
New AI models enhance hyperspectral image analysis for classification and super-resolution
Researchers have developed several new deep learning models for hyperspectral image analysis. The Dual-stage Spectrum-Constrained Clustering-based Classifier (DSCC) framework aims to improve classification accuracy by d…
-
一种基于图增强知识蒸馏的双流视觉Transformer结合区域感知注意力用于胃肠道疾病分类及可解释AI
研究人员开发了一种新颖的双流深度学习框架，用于从医学影像中对胃肠道疾病进行分类。该系统采用教师-学生知识蒸馏方法，结合了用于全局上下文的Swin Transformer和用于细粒度特征的Vision Transformer。学生网络是一个紧凑的Tiny-ViT，在数据集1上达到了0.9978的高准确率，在数据集2上达到了0.9928，AUC为1.0000，同时还提供了更快的推理速度和更低的计算复杂度。可解释性分析证实了该模型依赖于临床…
-
新方法改进了 AI 模型参数高效的多任务学习
研究人员开发了一种新的参数高效的多任务学习方法，用于计算机视觉。他们的方法称为渐进式任务特定适应，使用在早期层共享并在后期层变得更专业的适配器模块。这种设计有助于缓解在用有限的可训练参数将预训练模型适应多个任务时常见的任务干扰和负迁移问题。在 Swin 和 Pyramid Vision Transformers 上的评估表明，该方法在需要更少可训练参数的情况下优于现有的参数高效技术。
-
New methods QFlash and ELSA boost Vision Transformer attention efficiency
Researchers have developed two new methods to improve the efficiency of attention mechanisms in vision transformers. QFlash focuses on enabling integer-only operations for FlashAttention, achieving significant speedups …