From a904c16aa9910025bb93396dd2476b0603074712 Mon Sep 17 00:00:00 2001 From: ray4future <14845878+ray4future@user.noreply.gitee.com> Date: Tue, 21 Jan 2025 02:42:37 +0000 Subject: [PATCH] =?UTF-8?q?=E8=A7=84=E8=8C=83=E5=92=8C=E4=BD=8E=E9=94=99?= =?UTF-8?q?=E7=B1=BB=EF=BC=9A=20-=20=20=E9=94=99=E5=88=AB=E5=AD=97?= =?UTF-8?q?=E6=88=96=E6=8B=BC=E5=86=99=E9=94=99=E8=AF=AF=EF=BC=8C=E6=A0=87?= =?UTF-8?q?=E7=82=B9=E7=AC=A6=E5=8F=B7=E4=BD=BF=E7=94=A8=E9=94=99=E8=AF=AF?= =?UTF-8?q?=E3=80=81=E5=85=AC=E5=BC=8F=E9=94=99=E8=AF=AF=E6=88=96=E6=98=BE?= =?UTF-8?q?=E7=A4=BA=E5=BC=82=E5=B8=B8=EF=BC=9B=20-=20=20=E9=93=BE?= =?UTF-8?q?=E6=8E=A5=E9=94=99=E8=AF=AF=E3=80=81=E7=A9=BA=E5=8D=95=E5=85=83?= =?UTF-8?q?=E6=A0=BC=E3=80=81=E6=A0=BC=E5=BC=8F=E9=94=99=E8=AF=AF=EF=BC=9B?= =?UTF-8?q?=20-=20=20=E8=8B=B1=E6=96=87=E4=B8=AD=E5=8C=85=E5=90=AB?= =?UTF-8?q?=E4=B8=AD=E6=96=87=E5=AD=97=E7=AC=A6=EF=BC=9B=20-=20=20?= =?UTF-8?q?=E7=95=8C=E9=9D=A2=E5=92=8C=E6=8F=8F=E8=BF=B0=E4=B8=8D=E4=B8=80?= =?UTF-8?q?=E8=87=B4=EF=BC=8C=E4=BD=86=E4=B8=8D=E5=BD=B1=E5=93=8D=E6=93=8D?= =?UTF-8?q?=E4=BD=9C=EF=BC=9B=20-=20=20=E8=A1=A8=E8=BF=B0=E4=B8=8D?= =?UTF-8?q?=E9=80=9A=E9=A1=BA=EF=BC=8C=E4=BD=86=E4=B8=8D=E5=BD=B1=E5=93=8D?= =?UTF-8?q?=E7=90=86=E8=A7=A3=EF=BC=9B=20-=20=20=E7=89=88=E6=9C=AC?= =?UTF-8?q?=E5=8F=B7=E4=B8=8D=E5=8C=B9=E9=85=8D=EF=BC=9A=E5=A6=82=E8=BD=AF?= =?UTF-8?q?=E4=BB=B6=E5=8C=85=E5=90=8D=E7=A7=B0=E3=80=81=E7=95=8C=E9=9D=A2?= =?UTF-8?q?=E7=89=88=E6=9C=AC=E5=8F=B7=EF=BC=9B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: ray4future <14845878+ray4future@user.noreply.gitee.com> --- tutorials/source_zh_cn/cv/vit.ipynb | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tutorials/source_zh_cn/cv/vit.ipynb b/tutorials/source_zh_cn/cv/vit.ipynb index 59ae7ce4fd..cf349d7c41 100644 --- a/tutorials/source_zh_cn/cv/vit.ipynb +++ b/tutorials/source_zh_cn/cv/vit.ipynb @@ -445,7 +445,7 @@ "\n", "在ViT模型中：\n", "\n", - "1. 通过将输入图像在每个channel上划分成大小为16 x 16的patch，这一步是通过卷积操作来完成的，当然也可以人工进行划分，但卷积操作也可以达到目的同时还可以进行一次而外的数据处理；**例如一幅输入224 x 224的图像，首先经过卷积处理得到14 x 14个patch，那么每一个patch的大小就是16 x 16。**\n", + "1. 通过将输入图像在每个channel上划分成大小为16 x 16的patch，这一步是通过卷积操作来完成的，当然也可以人工进行划分，但卷积操作也可以达到目的同时还可以进行一次额外的数据处理；**例如一幅输入224 x 224的图像，首先经过卷积处理得到14 x 14个patch，那么每一个patch的大小就是16 x 16。**\n", "\n", "2. 再将每一个patch的矩阵拉伸成为一个一维向量，从而获得了近似词向量堆叠的效果。**上一步得到的一系列大小为16 x 16的patch就转换为长度为196的向量。**\n", "\n", -- Gitee