Python+Pytorch掌纹训练识别

程序示例精选
Python+Pytorch掌纹训练识别
如需安装运行环境或远程调试，见文章底部个人QQ名片，由专业技术人员远程协助！

前言

这篇博客针对《Python+Pytorch掌纹训练识别》编写代码，代码整洁，规则，易读。 学习与应用推荐首选。

文章目录

一、所需工具软件
二、使用步骤
       1. 主要代码
       2. 运行结果
三、在线协助

一、所需工具软件

       1. Python
       2. Pycharm

二、使用步骤

代码如下（示例）：

import torch 
from torch.utils.data import Dataset, DataLoader
import numpy as np
from models import MyDataset
from models import compnet
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print('\ndevice-> ', device, '\n\n')
test_set = './data/test.txt'
testset =MyDataset(txt=test_set, transforms=None, train=False)
batch_size = 1
data_loader_test = DataLoader(dataset=testset, batch_size=batch_size, shuffle=False)
net = compnet(num_classes=600) # IITD: 460    KTU: 145    Tongji: 600    REST: 358    XJTU: 200
#net.load_state_dict(torch.load('./pretrained models/net_params.pth'))
net.load_state_dict(torch.load('./pretrained models/net_params_best.pth'))
net.to(device)
net.eval()
# feature extraction:
featDB_test = []
iddb_test = []
with torch.no_grad():
    for batch_id, (data, target) in enumerate(data_loader_test):
        data = data.to(device)
        target = target.to(device)
        # feature extraction
        codes = net.getFeatureCode(data) 
        codes = codes.cpu().detach().numpy()
        y = target.cpu().detach().numpy()
        if batch_id == 0:
            featDB_test = codes
            iddb_test =  y
        else:
            featDB_test = np.concatenate((featDB_test, codes), axis=0)
            iddb_test = np.concatenate((iddb_test, y))
print('completed feature extraction for test set.')
print('(number of samples, feature vector dimensionality): ', featDB_test.shape)
print('\n')
# feature matching: feat1 vs feat2
cosdis =np.dot(feat1,feat2)
dis = np.arccos(np.clip(cosdis, -1, 1))/np.pi # 0~1
print('matching distance, label1 vs label2: \t%.2f, %d vs %d'%(dis, iddb_test[0], iddb_test[1]))

# feature matching: feat1 vs feat3 
cosdis =np.dot(feat1,feat3)
dis = np.arccos(np.clip(cosdis, -1, 1))/np.pi

print('matching distance, label1 vs label3: \t%.2f, %d vs %d'%(dis, iddb_test[0], iddb_test[-1]))
# Match label1 with all other samples
# Match label1 with all other samples
feat1 = featDB_test[0]  # Feature vector of label1
label1 = iddb_test[0]   # Label of the first sample
for i in range(1, len(featDB_test)):
    feat_other = featDB_test[i]  # Feature vector of the other sample
    label_other = iddb_test[i]   # Label of the other sample
    
    # Compute cosine similarity and convert to distance
    cosdis = np.dot(feat1, feat_other)
    dis = np.arccos(np.clip(cosdis, -1, 1)) / np.pi  # Distance in [0, 1]
   

运行结果

三、在线协助：

如需安装运行环境或远程调试，见文章底部个人 QQ 名片，由专业技术人员远程协助！