torch.nn.functional.grid_sample实验

最新推荐文章于 2024-06-20 19:31:58 发布

生成滞涨网络~

最新推荐文章于 2024-06-20 19:31:58 发布

阅读量255

点赞数

分类专栏： python 文章标签： python pytorch

本文链接：https://blog.csdn.net/qq_42061298/article/details/119699153

版权

PyTorch grid_sample 双线性插值坐标变换深度学习

关键词由CSDN通过智能技术生成

python 专栏收录该内容

22 篇文章 0 订阅

订阅专栏

# -*- coding: utf-8 -*-
# @Author: H.Yang
# @Date:   2021-08-14 11:58:04
# @Last Modified by:   H.Yang
# @Last Modified time: 2021-08-14 11:59:27
import torch
import torch.nn as nn

test = torch.rand(1,1,3,3)
test[0][0][0][0]=1
test[0][0][0][1]=2
test[0][0][0][2]=3
test[0][0][1][0]=4
test[0][0][1][1]=5
test[0][0][1][2]=6
test[0][0][2][0]=7
test[0][0][2][1]=8
test[0][0][2][2]=9
print(test)
#tensor([[[[1., 2., 3.],
#          [4., 5., 6.],
#          [7., 8., 9.]]]])

sample_one = torch.zeros(1,1,1,2)
sample_one [0][0][0][0] = -1  # x
sample_one [0][0][0][1] = -1  # y

sample_two = torch.zeros(1,1,1,2)
sample_two [0][0][0][0] = -2/3  # x
sample_two [0][0][0][1] = -2/3  # y

sample_thr = torch.zeros(1,1,1,2)
sample_thr [0][0][0][0] = -0.8  # x
sample_thr [0][0][0][1] = -0.3  # y

result_one = torch.nn.functional.grid_sample(test,sample_one,mode='bilinear',padding_mode="zeros",align_corners=True)
print(result_one )
#tensor([[[[1.]]]])

result_two= torch.nn.functional.grid_sample(test,sample_two ,mode='bilinear',padding_mode="zeros",align_corners=False)
print(result_two)
#tensor([[[[1.]]]])

result_thr= torch.nn.functional.grid_sample(test,sample_thr ,mode='bilinear',padding_mode="zeros",align_corners=True)
print(result_thr)
#tensor([[[[3.3000]]]]) 通过双线性插值，坐标距离越近权重越大，直接算就可以得到(1*0.8+2*0.2)*0.3+(4*0.8+5*0.2)*0.7=3