- 博客(19)
- 收藏
- 关注
原创 pytorch图像检索评价指标MAP
map是图像检索模型的一个评价指标以图片中第一个计算AP值为例P的分别是[1, 2/3, 3/6, 4/9, 5/10]R值分别是[1/5, 2/5, 3/5/, 4/5, 1]AP计算结果(1+2/3+3/6+4/9+5/10)/ 5https://blog.csdn.net/weixin_40100431/article/details/88768479假设输入一张图片,数据库中有3个图像与之匹配,按照相似度高低进行排序,得到的位置是第1位,第3位,第6位是这三幅图像,匹配正确
2021-04-26 10:44:49 1815
原创 基于resnet18的自然图像检索
# -*- encoding: utf-8 -*-"""@File : 基于resnet18的自然图像检索.py@Time : 2021-04-18 14:25@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""from tqdm import tqdmimport globfrom sklearn.preprocessing import normalizeimport matplotlib.pyplo
2021-04-20 16:07:11 831 1
原创 基于颜色直方图的图像检索
使用数据第一个是有1000张图像的自然图像数据集第二个是牛津大学建筑数据集# -*- encoding: utf-8 -*-"""@File : 颜色直方图图像检索.py@Time : 2021-04-18 10:56@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""from tqdm import tqdmimport globfrom sklearn.preprocessing import
2021-04-20 16:03:04 691 1
原创 python绘制双正切激活函数
import numpy as npimport matplotlib.pyplot as plt#f(x) = (e**x - e ** -x) / (e**x + e ** -x)#绘制此函数x = np.arange(-5,5,0.1)y = (np.exp(x) - np.exp(-x)) / (np.exp(x) + np.exp(-x))print(len(x))print(len(y))plt.plot(x,y)plt.show()
2021-04-20 09:43:06 199
原创 minist利用训练好的model推断并且可视化
import torchfrom torchvision import transformsfrom torchvision import datasetsimport torchvision as tvfrom torch.utils.data import DataLoaderimport torch.nn.functional as Fimport torch.optim as optimimport osimport numpy as npbatch_size = 64t
2021-04-15 22:08:26 117
原创 tv.utils.make_grid
import torchfrom torchvision import transformsfrom torchvision import datasetsfrom torch.utils.data import DataLoaderfrom torchvision.transforms import ToPILImageimport torch.nn.functional as Fimport torch.optim as optimimport torch.nn as nnimport
2021-04-15 16:55:25 334
原创 ToPILImage
import matplotlib.pyplot as pltimport numpy as npshow = ToPILImage()data = torch.rand(3,64,128)data = show(data)plt.imshow(data)plt.show()
2021-04-15 16:18:41 2308 1
原创 pytorch网络可学习参数
import torchfrom torchvision import transformsfrom torchvision import datasetsfrom torch.utils.data import DataLoaderimport torch.nn.functional as Fimport torch.optim as optimbatch_size = 64transform = transforms.Compose([ transforms.ToTensor(
2021-04-15 10:22:51 561
原创 反向传播梯度的计算
import torch as tfrom torch.autograd import Variable as Vdef f(x): # 计算输出值 y = x ** 2 * t.exp(x) return ydef gradf(x): # 手动计算梯度 dx = 2 * x * t.exp(x) + x ** 2 * t.exp(x) return dxx = V(t.randn(3, 4), requires_grad=True)y = f(x
2021-04-14 23:12:46 724
原创 查看tensor的形状,行列大小
torch.size是tuple(元组的子类,因此他支持元组所有的操作,如x.size()[0]import pandas as pdimport numpy as npimport torchx = torch.rand(5,3)print(x.size())print(x.size()[0])print(x.size()[1])torch.Size([5, 3])53
2021-04-14 22:42:13 6590
原创 one hot 编码
方式1import torchindex = [0,1,2,3,4,5,6,7,8,9]onehot = pd.get_dummies(index)onehot = np.array(onehot)onehot = torch.tensor(onehot)print(onehot)方式2import numpy as npimport torchindex = torch.tensor([0,1,2,3,4,5,6,7,8,9])t = torch.zeros(10,10)
2021-04-13 23:06:46 147
原创 pytorch18种损失函数全详解
https://blog.csdn.net/nstarlds/article/details/104733459/
2021-04-12 22:47:35 292
原创 cbir
核心问题1需要用什么样的特征去表征图像网络hog,sift,gist基于具体的场景目标去训练卷积神经网络提取神经网络生成输出作为特征2如何定义相似各种距离准则3高维空间里面如何高效的查找最近邻ANN算法大有帮助卷积神经网训练的同时学习分桶桶号hog特征抽取0605年cvpr预先划分数据ANN库FLANN库...
2021-04-10 14:39:30 142
原创 argparse库的使用
import argparseparser = argparse.ArgumentParser(description='Process some integers.')parser.add_argument('integers', metavar='N', type=int, nargs='+', help='an integer for the accumulator')parser.add_argument('--sum', dest='accumula
2021-04-09 10:11:18 82
原创 python 工程生成requirements
https://blog.csdn.net/u014439564/article/details/99624361?ops_request_misc=&request_id=&biz_id=102&utm_term=pipreqs&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduweb~default-6-99624361.pc_search_result_no_baidu_js
2021-04-08 10:29:05 141
原创 MINIST 数据展示代码可视化minist(深入浅出pytorch)
import torchfrom torchvision import transformsfrom torchvision import datasetsfrom torch.utils.data import DataLoaderimport torch.nn.functional as Fimport torch.optim as optimimport torch.nn as nnimport numpy as npbatch_size = 64transform = tran
2021-04-07 22:14:34 332
原创 kaggle猫狗识别pytorch
https://zhuanlan.zhihu.com/p/136421422数据集下载链接:https://pan.baidu.com/s/1KZlADFFgI1zd_ia10U5FWw 提取码:r7wn 复制这段内容后打开百度网盘手机App,操作更方便哦
2021-04-06 15:09:43 467
原创 计算fc层超参数
全连接层参数1408是一个超参数,需要人工计算。可以再forward中先不写全连接层构建一个tensor输入然后看一下输出x.size()就是全连接层的参数
2021-04-01 11:33:14 586
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人