孪生网络代码改动2（增加自动打标与准确率验证模块）

2.1主模块 ：

only show accuracy:如果输入 y ，仅验证分类结果的最准确率，默认标签地址为output/mac(DAGM)/2022_05_01 01_51_04，均可自行调整

[0]model path: 默认地址model path: logs/mac_5/ep098-loss0.000-val_loss0.002.pth，注意加载网络权重后，记得调整siamese.py中的input_shape

[1]待检测图片文件夹(query set)地址：
[2]待检测图片，support set 地址：默认地址为query/mac_mini
[3]标签图片，support/mac(DAGM)：默认地址为support/mac(DAGM)

# -*- coding:utf-8 -*-
import datetime
from utils.function import *
from PIL import Image
from siamese import Siamese

# 实例化网络
# Model_path = input(f'model weights:')
# if Model_path == '':
#     Model_path = rf'logs/animal300_200/ep039-loss0.038-val_loss0.251.pth'
# print(f'1.模型权重:{Model_path}')
model = Siamese()
# model.generate(Model_path)

# 对比（相似度计算）函数
#相似度函数

def TwoImg(img1, img2):
    try:
        img = Image.open(img1)
    except:
        print('Image Open Error! Try again!')
    try:
        img2 = Image.open(img2)
    except:
        print('Image_2 Open Error! Try again!')

    # 此时img已经转化为PIL格式
    probability = model.detect_image(img, img2)  # 主要函数（孪生网络模型函数）
    # print(probability)
    return probability

# 识别函数
def recognition(Img, List2):
    # 初始化
    prob = []
    label = []
    result = {}
    # 与每张标签图片进行对比，返回相似度
    for Path in List2:
        path = Path.split('\\')[-1]
        first, last = os.path.splitext(path)
        img2 = Path  # 标签图片
        # 对比，输入图片地址，返回相似度（tensor格式）
        prob_two = TwoImg(Img, img2)
        # 将相似度tensor格式（prob_two）转化为float格式（similarity）
        similarity = prob_two.tolist()[0]
        # 将相似度结果加入相似度列表（prob）
        prob.append(similarity)
        # 将图片名称加入标签列表（label）
        label.append(first)

    # 打印相似度和图片名称列表
    # print(f'\n1.标签为：{label}\n'
    #       f'2.与待检测图片相似度分别为：{prob}')

    # 对相似度列表进行softmax转换
    prob_sm = softmax(prob)
    # print(f'3.各个类别的可能性：{prob_sm}\n')

    # 打印各个类别的可能性百分比
    for i in range(len(label)):
        print(f'{label[i]}: {prob_sm[i] * 100:.2f}%(相似度：{prob[i] * 100:.5f}%)')
        # 编制result字典：1.key为标签 label[i] 2.value为经softmax的相似度（即概率）prob_sm[i]
        result[f'{label[i]}'] = prob_sm[i]

    # 最有可能的预测结果
    MAX = sorted(result,
                 key=result.get,
                 reverse=True)[0]

    print(f'\n{Img}最有可能的结果是【{MAX}】,'
          f'有{result[MAX] * 100:.2f}%的可能性.\n')
    return prob_sm, result, MAX

#主函数
def main_sn():
    if not input(f'1.only show accuracy:') == 'y':
        # TODO 1.待检测图片地址
        img_path1 = input(f'待检测图片文件夹(query set)地址：')  # 待检测图片文件夹地址
        if img_path1 == '':
            img_path1 = rf'query/mac_mini'
        print(f'2.待检测图片:{img_path1}')
        list1 = GetFileList(img_path1, [])

        # TODO 标签图片（label images）地址
        img_path2 = input(f'support set 地址：')
        if img_path2 == '':
            img_path2 = rf'support/mac(DAGM)'
        print(f'3.标签图片:{img_path2}')
        # 获取标签图片名称等信息
        list2 = GetFileList(img_path2, [])

        num = 1
        # 保存文件夹名称
        define = '%Y_%m_%d %H_%M_%S'
        system_time = datetime.datetime.now().strftime(define)
        supportSet = img_path2.split('/')[-1]
        save_path = rf'output/{supportSet}/'
        for img in list1:
            # img为待检测图片地址
            print(f'-----------------------------------------image[{num}]----------------------------------------')
            prob_sm, result, Max = recognition(img, list2)
            if not os.path.exists(f'{save_path}/{system_time}/labels'):
                os.makedirs(f'{save_path}/{system_time}/labels')
            with open(rf'{save_path}/{system_time}/info.txt', 'a+') as f:
                f.write(f'{Max} {result}\n')
            first_img, last_img = img.split('\\')
            last_img_excludePNG = last_img.split('.')[:-1][0]
            with open(rf'{save_path}/{system_time}/labels/{last_img_excludePNG}.txt', 'a+') as f:
                f.write(f'{Max}')
            num += 1
        predict_accuracy(rf'{save_path}/{system_time}')
        return 'OK'
    predict_accuracy(rf'output/mac(DAGM)/2022_05_01 01_51_04')

if __name__ == "__main__":
    main_sn()

2.2函数模块

import os
import numpy as np
import os.path


# 输入文件夹地址和[]，返回包含子文件名称的列表
def GetFileList(dir, fileList):
    newDir = dir
    if os.path.isfile(dir):
        fileList.append(dir)
    elif os.path.isdir(dir):
        for s in os.listdir(dir):
            # 如果需要忽略某些文件夹，使用以下代码
            # if s == "xxx":
            # continue
            newDir = os.path.join(dir, s)
            GetFileList(newDir, fileList)
    return fileList


# softmax转换 输入输出为列表
def softmax(x):
    """Compute the softmax in a numerically stable way."""
    x = x - np.max(x)
    exp_x = np.exp(x)
    softmax_x = exp_x / np.sum(exp_x)
    return softmax_x

def predict_accuracy(result_path):
    fileList = []
    result_path = rf'{result_path}\labels'
    filelist = GetFileList(result_path, fileList)
    total = len(filelist)
    num = 0
    acc = 0
    for txt in filelist:
        txtPath = txt.split('\\')[-1:][0]
        class_num, number_classes = txtPath.split('_')
        if not os.path.exists(txt):
            print(f'txt文件地址不存在')
        with open(txt, 'r') as f:
            predict_result = f.readline()
        if predict_result == class_num[5:]:
            acc += 1
        num += 1
        print(f'[{num}]预测准确率：{float(acc/num)*100:.2f}%\n进度：{float(num/total)*100:.2f}%')
    with open(rf'{result_path}/predict_accuracy.txt', 'a+') as f:
        f.write(f'{predict_accuracy}')

def demo(path):
    list = []
    for line in open(path):
        line_first = str(line.split('\\')[1])
        print(line_first)
        list.append(line_first)
    f = open(path,'w')
    for i in range(len(list)):
        f.write(list[i])


if __name__ == '__main__':
    # 测试GetFileList函数
    list = GetFileList('../support/text', [])
    for path in list:
        print(path)
    predict_accuracy(rf'../output/mac(DAGM)/2022_05_01 01_51_04')

    #demo(r'C:\Users\cleste\Desktop\孪生网络（少样本分类）\demo.txt')

2.3 针对siamese.py模块的修改

siamese.py模块中增加了如下语句：控制打印的图像数量
if float(output) > 0.01: # 置信度对于1%的图片对才会被打印