Kaggle on 南巷https://nx.wmlab.top/tags/kaggle/Recent content in Kaggle on 南巷Hugo -- gohugo.iozh-CNliuwenhao1968@163.com (南巷)liuwenhao1968@163.com (南巷)© 2026 南巷Sat, 01 Feb 2025 00:00:00 +0000Kaggle 树叶图像分类https://nx.wmlab.top/competitions/kaggle-leaf-competition/Sat, 01 Feb 2025 00:00:00 +0000liuwenhao1968@163.com (南巷)https://nx.wmlab.top/competitions/kaggle-leaf-competition/基于 ResNet50 优化细粒度特征表达，结合 Label Smoothing、Focal Loss 与 AdamW + CosineAnnealing 策略。2018 Data Science Bowl：细胞核分割https://nx.wmlab.top/portfolio/data-science-bowl-nuclei-segmentation/Wed, 15 Jan 2025 00:00:00 +0000liuwenhao1968@163.com (南巷)https://nx.wmlab.top/portfolio/data-science-bowl-nuclei-segmentation/改进 U-Net 进行细胞核实例分割，结合 Deformable Convolution、密集跳跃连接和强数据增强，最终排名前 5%。2018 Data Science Bowlhttps://nx.wmlab.top/competitions/data-science-bowl-2018/Wed, 01 Jan 2025 00:00:00 +0000liuwenhao1968@163.com (南巷)https://nx.wmlab.top/competitions/data-science-bowl-2018/改进 U-Net，引入 Deformable Convolution、密集跳跃连接与复合损失，提升核形态和尺度建模能力。