1 #加载需要的库文件 2 import torch 3 import torch.nn as nn 4 #输入Tensor : A = [batch_size,channels，width,height] 5 A = torch.rand(16,3,512,512) # 表示16张行列号均为512的彩色图像 6 #获取单张图像shape--->[channels，width,height]: 7 img_shape = A[0].shape 8 print("img_shape : ",img_shape) 9 #获取图像的行列号--->[width,height]: 10 img_Size = A[0][0].shape 11 print("img_Size : ",img_Size) 12 #获取batch内第0个通道数据,冒号代表选择全部: 13 zero_imgs = A[:,0,:,:] 14 print("zero_imgs shape : ",zero_imgs.shape) 15 #获取batch内属于第0和第1通道数据,冒号代表选择全部 16 batch_two = A[:,:2,:,:] 17 print("batch_two shape : ",batch_two.shape) 18 #获取batch内属于第3通道的数据 19 batch_last = A[:,-1:,:,:] #负号表示从后向前获取 20 batch_Last = A[:,2:,:,:] 21 print("batch_last shape : ",batch_last.shape) 22 print("batch_Last shape : ",batch_Last.shape) 23 #获取batch内前5张图像 24 batch_img = A[:5,:,:,:] 25 print("batch_img : ",batch_img.shape) 26 #以步长为2获取batch内的图像数据 27 batch_data = A[:,:,0:512:2,0:512:2] 28 batch_Data = A[:,:,::2,::2] 29 print("batch_data shape : ",batch_data.shape) 30 print("batch_Data shape : ",batch_Data.shape)
结果为：
1 img_shape : torch.Size([3, 512, 512]) 2 img_Size : torch.Size([512, 512]) 3 zero_imgs shape : torch.Size([16, 512, 512]) 4 batch_two shape : torch.Size([16, 2, 512, 512]) 5 batch_last shape : torch.Size([16, 1, 512, 512]) 6 batch_Last shape : torch.Size([16, 2, 512, 512]) 7 batch_img : torch.Size([5, 3, 512, 512]) 8 batch_data shape : torch.Size([16, 3, 256, 256]) 9 batch_Data shape : torch.Size([16, 3, 256, 256])
input ： 输入的Tensor
dim : 索引的轴
index ：索引的轴的索引
out ： 目标Tensor
1 A = torch.rand(3,4) 2 print(A) 3 #获取第一行，第三行数据 4 indices = torch.LongTensor([0,2]) 5 data = torch.index_select(A, 0, indices) 6 print(data) 7 #获取第三列数据 8 indices = torch.LongTensor([2]) 9 data = torch.index_select(A, 1, indices) 10 print(data)
结果为：
1 tensor([[0.3413, 0.4133, 0.8881, 0.3013], 2 [0.2296, 0.8172, 0.6642, 0.0631], 3 [0.0300, 0.5728, 0.0011, 0.6917]]) 4 5 tensor([[0.3413, 0.4133, 0.8881, 0.3013], 6 [0.0300, 0.5728, 0.0011, 0.6917]]) 7 8 tensor([[0.8881], 9 [0.6642], 10 [0.0011]])
1 A = torch.rand(3,4) 2 print(A) 3 mask = A.ge(0.5) 4 mask_data = torch.masked_select(A, mask) 5 print(mask_data)
结果为：
1 tensor([[0.3891, 0.6926, 0.3271, 0.9869], 2 [0.4256, 0.9135, 0.3202, 0.8631], 3 [0.2141, 0.6315, 0.4464, 0.4018]]) 4 5 tensor([0.6926, 0.9869, 0.9135, 0.8631, 0.6315])
inputs (sequence of Tensors)
dimension (int, optional) – 沿着此维连接张量序列。
在给定维度上对输入的张量序列 seq 进行连接操作
input (Tensor) – 源张量
out (LongTensor, optional) – 包含索引值的结果张量
tensor (Tensor) – 待分割张量
split_size (int) – 单个分块的形状大小
dim (int) – 沿着此维进行分割
input (Tensor) – 输入张量
dim (int, optional) – 如果给定，则input只会在给定维度挤压
out (Tensor, optional) – 输出张量
沿着一个新维度对输入张量序列进行连接。序列中所有的张量都应该为相同 形状
原文：https://www.cnblogs.com/yzj-notes/p/14743935.html
