知识卡片
图像展平必须固定像素顺序
28×28灰度图可按一致顺序展平成784维向量作为全连接网络输入。展平本身不丢像素数值,却不再显式利用二维局部共享结构;训练与推理展平顺序不一致会改变特征含义。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 深度学习模型
原书出处 · 1 处
- · 查看原书依据
We begin by considering each handwritten digit as a low-resolution 28 × 28 grayscale image. In computers, each image is stored as an array of numbers with each value denoting the pixel intensity, which varies between 0 and 255 in integer values. Thus, each image of these handwritten digits is seen by a computer as a 28 × 28 matrix (this process is visualized in
参考来源
- 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第08章《深度学习模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/21_chapter08.xhtml)
- 结论依据:摘录中关于图像展平必须固定像素顺序的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:We begin by considering each handwritten digit as a low-resolution 28 × 28 grayscale image. In computers, each image is stored as an array of numbers with each value denoting the pixel intensity, which varies between 0 and 255 in integer values. Thus, each image of these handwritten digits is seen by a computer as a 28 × 28 matrix (this process is visualized in