![](https://img-blog.csdnimg.cn/20201014180756925.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
pytorch学习
范德彪陕西分彪
Chosen one
展开
-
绘制2d_feature
2d_feature原创 2022-08-07 01:52:08 · 84 阅读 · 0 评论 -
face匹配指标
文章目录人脸匹配算法性能评价指标人脸验证人脸识别人脸匹配算法性能评价指标人脸验证ROC曲线PR曲线人脸识别CMC曲线原创 2022-01-15 22:05:43 · 1502 阅读 · 0 评论 -
根据x-ray生成训练csv文件
import osimport pandas as pdl_nor = os.listdir(train_nor_dir)l_nor_label = [0 for i in l_nor]l_pne = os.listdir(train_pne_dir)l_pne_label = [1 for i in l_pne]l_img = l_nor + l_pnel_label = l_nor_label + l_pne_labeldata = {"img": l_img, "l_labe原创 2022-01-14 00:02:09 · 2104 阅读 · 0 评论 -
将id->label
df = pd.read_csv(csv_train_path)df.head()unique_classes = pd.unique(df['Id'])encoding = dict(enumerate(unique_classes))encoding = {value: key for key, value in encoding.items()}df = df.replace(encoding)unique_classes = pd.unique(df['Id'])encoding原创 2022-01-13 23:30:27 · 220 阅读 · 0 评论 -
df.values
train_df.values[:10]array([['0000e88ab.jp g', 'w_f48451c'], ['0001f9222.jpg', 'w_c3d896a'], ['00029d126.jpg', 'w_20df2c5'], ['00050a15a.jpg', 'new_whale'], ['0005c1ef8.jpg', 'new_whale'], ['0006e997e.jpg', 'new_whale'],原创 2022-01-13 23:16:21 · 2492 阅读 · 0 评论 -
不同类别的图像数量可视化
文章目录鲸鱼数据可视化chest-xray鲸鱼数据https://www.kaggle.com/mmm4a1/seresnext101-pytorch-starter/edit利用train.csv对其不同类别的图像数量进行可视化。首先读取一下train_df = pd.read_csv('../input/train.csv')看一下数据的样子,我们要用到Id这一列。train_df.head()看一下一共有多少种鲸鱼图像?NCLASSES = len(train_df['Id'原创 2022-01-13 22:15:27 · 295 阅读 · 0 评论 -
类激活热图
文章目录CAMgrad camcam方法实践讲了一大堆,并没有听懂,直接看代码吧。。grad-cam是cam方法的一种泛化。CAM需要全局池化gapCAM图必须用全局平均池化。grad camcam方法实践讲了一大堆,并没有听懂,直接看代码吧。。grad-cam是cam方法的一种泛化。ps:有三ai课不太行。。...原创 2022-01-07 20:58:06 · 531 阅读 · 0 评论 -
attention
文章目录基础网络结构facenetpyTorch实现注意力机制SENET实现方式pyTorch实现CBAM实现方式pyTorch实现ECA实现方式pyTorch实现NEXT模型融合CAM可视化2Params size (MB): 0.00Estimated Total Size (MB): 2.66基础网络结构facenet主干网络:resnet50损失函数:nn.CrossEntropyLoss() + 三元组损失pyTorch实现facenetimport torchfr原创 2021-12-30 20:45:53 · 999 阅读 · 0 评论 -
CV注意力机制
Senet# -*- encoding: utf-8 -*-"""@File : seNet.py@Time : 2021-12-29 17:44@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""import torchfrom torch import nnfrom torchsummary import summaryclass seNet(nn.Module): def __init__原创 2021-12-29 22:01:37 · 637 阅读 · 0 评论 -
mobilenet V1
https://blog.csdn.net/baidu_36913330/article/details/120079049# 非官方实现import torchclass MobileNet(nn.Module): def __init__(self): super(MobileNet, self).__init__() def conv_bn(inp, oup, stride): # 第一层传统的卷积:conv3*3+BN+ReLU .原创 2021-12-13 15:24:00 · 973 阅读 · 0 评论 -
pytorch得到中间层输出
讲解视频https://www.bilibili.com/video/BV1z7411f7za?spm_id_from=333.999.0.0关键代码import torch.nn as nnimport torchclass AlexNet(nn.Module): def __init__(self, num_classes=1000, init_weights=False): super(AlexNet, self).__init__() self原创 2021-12-08 21:24:49 · 811 阅读 · 0 评论 -
跨领域掌纹识别
http://tlo.xjtu.edu.cn/info/1012/4456.htm原创 2021-11-20 23:19:55 · 240 阅读 · 0 评论 -
提高CIFAR-10分类准确度的方法
https://www.zhihu.com/question/452410923原创 2021-11-18 20:03:04 · 1149 阅读 · 0 评论 -
孪生网络Dataset
以natureDB为例natureDB是一个自然图像数据集,一共有十个类别一个类别有100张照片。利用脚本将图像的类别和和路径存入txt文件(这个很容易根据情况写就行了。。).class nature_dataset(): def __init__(self, transform = None, train = True): """ :param transform: 选择transform:transform_train/test_train原创 2021-11-17 22:35:53 · 488 阅读 · 0 评论 -
原始的DSH深度哈希代码
https://github.com/weixu000/DSH-pytorch/blob/master/main.py原创 2021-11-15 08:29:55 · 326 阅读 · 0 评论 -
腾讯云blog:孪生网络入门(上) Siamese Net及其损失函数
腾讯云平台有一些很好地文章值得一看https://cloud.tencent.com/developer/article/1761814原创 2021-11-14 16:24:33 · 361 阅读 · 0 评论 -
One Shot Learning with Siamese Networks
https://github.com/harveyslash/Facial-Similarity-with-Siamese-Networks-in-Pytorch/blob/master/Siamese-networks-medium.ipynb原创 2021-11-14 14:28:28 · 2441 阅读 · 0 评论 -
deephash项目代码使用指北
指南PASS:Microsoft Windows [版本 10.0.19043.1288](c) Microsoft Corporation。保留所有权利。(pytorch) E:\xd\DeepHash-pytorch-master\numpy-main>cd .(pytorch) E:\xd\DeepHash-pytorch-master\numpy-main>cd ~系统找不到指定的路径。(pytorch) E:\xd\DeepHash-pytorch-master\n原创 2021-11-01 19:26:07 · 102 阅读 · 0 评论 -
map计算
import numpy as npdata_feature = np.random.rand(10000,512)data_lable = np.random.randint(low = 1, high = 10, size = 10000)#print(data_lable)query = np.random.rand(1,512)query_label = 1# M = np.dot(data_feature,query.T)# rank = M.flatten().argsort()[原创 2021-10-29 22:38:36 · 255 阅读 · 0 评论 -
maketrainlabel
# -*- encoding: utf-8 -*-"""@File : make_traintxt.py@Time : 2021-10-28 12:31@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""import torchimport globimport os.path as ospimport randomdata_folder = "G:\data\cifar10"im_train_lis原创 2021-10-28 15:51:18 · 107 阅读 · 0 评论 -
STN可以增强CNN网络
https://pytorch.org/tutorials/intermediate/spatial_transformer_tutorial.html原创 2021-10-27 16:36:52 · 358 阅读 · 0 评论 -
torch.round()
>>> a = torch.randn(4)>>> atensor([ 0.9920, 0.6077, 0.9734, -1.0362])>>> torch.round(a)tensor([ 1., 1., 1., -1.])原创 2021-10-27 11:08:50 · 4400 阅读 · 0 评论 -
numpy存储特征:消耗内存和计算速度测试
创建一个10w * 512 的浮点数矩阵,和一个10 * 512 的int8矩阵。import numpy as npm = np.random.rand(100000, 512)np.save("data/feature/512.npy", m)n = np.ones((3, 3),dtype = "int8")np.save("data/feature/hash512.npy", n)首先测试一下汉明距离计算速度矩阵大小:100000 * 512feature:1 * 512i原创 2021-10-24 15:05:23 · 318 阅读 · 0 评论 -
arcface和cosface
文章中的图解决了我长久以来的疑惑https://blog.csdn.net/qq_34914551/article/details/104522030原创 2021-09-04 15:14:12 · 350 阅读 · 0 评论 -
pytorch tensorboard
TORCH.UTILS.TENSORBOARDBefore going further, more details on TensorBoard can be found at https://www.tensorflow.org/tensorboard/Once you’ve installed TensorBoard, these utilities let you log PyTorch models and metrics into a directory for visualization w原创 2021-08-11 11:35:12 · 171 阅读 · 0 评论 -
zwjt题
B# -*- encoding: utf-8 -*-"""@File : 666.py@Time : 2021-08-10 19:07@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""import torchfrom torch import nninput = torch.randn(1,1,100,100)class Net(nn.Module): def __init__(self)原创 2021-08-10 20:47:18 · 170 阅读 · 0 评论 -
Deep Learning of Binary Hash Codes for Fast Image Retrieval(代码跑通了)
deephash:项目地址(pytorch14) liuyitao@gpu3:~/XIAN_AI/xuda/deephash/pytorch_deephash-master$ python train.pyFiles already downloaded and verifiedFiles already downloaded and verifiedUse device: cuda:0Epoch: 1Training: 100%|███████████████████████████████原创 2021-07-22 14:34:19 · 194 阅读 · 4 评论 -
PyTorch中的model.modules(), model.children(), model.named_children(), model.parameters(), model.nam...
https://www.jianshu.com/p/a4c745b6ea9b原创 2021-07-21 16:09:48 · 303 阅读 · 0 评论 -
resnet50结构
D:\ANACONDA\envs\pytorch_gpu\python.exe G:/各种代码/proj/proj_708/NatureNet.pyPython 3.6.12 |Anaconda, Inc.| (default, Sep 9 2020, 00:29:25) [MSC v.1916 64 bit (AMD64)]Type 'copyright', 'credits' or 'license' for more informationIPython 7.16.1 -- An enhanc转载 2021-07-20 17:54:45 · 142 阅读 · 0 评论 -
各种池化操作(包括组合池化)
https://www.cnblogs.com/pprp/p/12456403.html原创 2021-07-20 17:30:07 · 613 阅读 · 0 评论 -
关于类ImageFolder
值得一看http://www.bubuko.com/infodetail-3540954.html原创 2021-07-16 17:39:14 · 93 阅读 · 0 评论 -
DCGAN baseline
DCGAN baseline我们一共需要两个脚本model.py 写生成器,判别器和初始化权重函数train脚本用于训练网络并生成结果首先看model.py文件引入相关包import torchimport torch.nn as nn鉴别器Discriminatorclass Discriminator(nn.Module): def __init__(self, channels_img, features_d): super(Discriminator,原创 2021-07-15 21:50:02 · 122 阅读 · 0 评论 -
VGG16结构
Sequential( (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (1): ReLU(inplace=True) (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (3): ReLU(inplace=True) (4): MaxPool2d(kernel_size=2, stride=2, paddi原创 2021-07-08 17:34:40 · 133 阅读 · 0 评论 -
20天掌握Pytorch文档链接
http://eat.woshinlper.com/原创 2021-06-16 01:38:28 · 171 阅读 · 0 评论 -
icemdata_manager.py
# -*- encoding: utf-8 -*-"""@File : 15616.py@Time : 2021-05-14 16:27@Author : XD@Email : gudianpai@qq.com@Software: PyCharm"""import os.path as ospimport globfrom IPython import embedclass ICME(object): dataset_dir = 'icme'原创 2021-05-14 22:55:01 · 73 阅读 · 0 评论 -
Aligned TripletLoss
class AlignedTripletLoss(nn.Module): def __init__(self, margin=0.3): super().__init__() # margin就是三元组损失中的边界α self.margin = margin # 计算三元组损失使用的函数 self.ranking_loss = nn.MarginRankingLoss(margin=margin) def fo原创 2021-06-08 21:14:55 · 72 阅读 · 0 评论 -
AlignedReID local_dist
import torchdef batch_euclidean_dist(x, y): """ Args: x: pytorch Variable, with shape [Batch size, Local part, Feature channel] y: pytorch Variable, with shape [Batch size, Local part, Feature channel] Returns: dist: pytorch Variable, .原创 2021-06-05 16:37:57 · 90 阅读 · 0 评论 -
AlignedReID 网络改写
from __future__ import absolute_importimport torchfrom torch import nnfrom torch.nn import functional as Fimport torchvisionfrom aligned.HorizontalMaxPool2D import HorizontalMaxPool2d__all__ = ['ResNet50', 'ResNet101']class ResNet50(nn.Module):.原创 2021-06-05 14:55:11 · 211 阅读 · 0 评论 -
特征图水平池化
codeimport torch.nn as nn#input ->N * C * H * Wclass HorizontalMaxPool2d(nn.Module): def __init__(self): super(HorizontalMaxPool2d, self).__init__() def forward(self, x): inp_size = x.size() return nn.functional.max_.原创 2021-06-05 14:00:44 · 500 阅读 · 0 评论 -
pytorch:admm
distance.addmm_(mat1 = inputs, mat2 = inputs.t(), beta = 1, alpha = -2)这句就是计算distance = mat1 * 1 + mat2 * -2此外注意adamm和adamm_的区别distance.adamm_()会把计算后的值赋给distance而adamm不会原创 2021-05-29 15:23:19 · 833 阅读 · 0 评论