AIHIA梦工厂

标题: Insightface记录——验证集制作 [打印本页]

作者: 骑着拖拉机 时间: 2023-9-24 05:32 PM
标题: Insightface记录——验证集制作
**一、pairs 文件制作**
名单包含两种内容：

标签为1，是同一个人的两两组合图相对，
不同人的两两组合图像对，标签是0，代表是不同人

![在这里插入图片描述]( )

看网上很多人写的在生成不同人的图像对时忽略了，同一个人员ID目录下包含多张图的情况，我这里进行了修改，代码如下：

```bash
import argparse
import itertools
import os
import random
import time

if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Generate image pairs.')
parser.add_argument('--dataset', default='G:/myFaceData/myvaldata/imgs/',
                     type=str, help='Dataset path.')
parser.add_argument('--output', default='G:/myFaceData/myvaldata/pairs1.txt', type=str, help='Output path.')
parser.add_argument('--same', default=5000, type=int,
                     help="Number of same pairs for each person.")
args = parser.parse_args()

assert args.dataset and args.output, \
      'Dataset and output should be defined.'
file = open(args.output, 'w+')
#file.writelines('\n')
allList = []
# ---- Generate same pairs for each person
same = 0
personList = os.listdir(args.dataset)
samePairs = []
for person in personList:
      if not os.path.isdir(os.path.join(args.dataset, person)):
         continue
      IDPath = os.path.join(args.dataset, person)
      subList = []
      images = os.listdir(os.path.join(args.dataset, person))
      for image in images:
         if not image.lower().endswith(('jpg', 'jpeg', 'png')):
            continue
         imagePath = os.path.join(args.dataset, person, image)

         allList.append(imagePath)#所有的路径都存下来
         subList.append(imagePath)


      for pair in itertools.combinations(subList, 2):
         samePairs.append(pair)

choices = random.sample(samePairs, args.same)
for pair in choices:
      file.writelines(f'{pair[0]} {pair[1]} {1}\n')
      same += 1
# ---- Generate different pairs
print(f"Will generate {same} different pairs.")
count = 0
sets = set()
while True:
      if count >= same:
         break
      random.seed(time.time() * 100000 % 10000)
      random.shuffle(allList)
      indexa=random.sample(range(0, len(allList)),1)[0]
      indexb=random.sample(range(0, len(allList)),1)[0]
      IDPath_1 = os.path.split(allList[indexa])
      IDPath_2 = os.path.split(allList[indexb])
      if allList[indexa] != allList[indexb] and IDPath_1[0]!= IDPath_2[0] and not (allList[indexa], allList[indexb]) in sets:
         sets.add((allList[indexa], allList[indexb]))
         sets.add((allList[indexb], allList[indexa]))
         file.writelines(f'{allList[indexa]} {allList[indexb]} {0}\n')
         count += 1
```
**二、bin 文件制作**

同样网上找到跑不通，自己修改了一些内容。

```bash
#coding:utf-8

import mxnet as mx
from mxnet import ndarray as nd
import argparse
import pickle
import sys
import os
import numpy as np
import pdb
import matplotlib.pyplot as plt

def read_pairs(pairs_filename):
pairs = []
with open(pairs_filename, 'r') as f:
      for line in f.readlines()[0:]:
         pair = line.strip().split(',')
         pairs.append(pair)
return np.array(pairs)

def get_paths(pairs, same_pairs):
nrof_skipped_pairs = 0
path_list = []
issame_list = []
cnt = 1
for pair in pairs:
      path0 = pair[0].split(' ')[0]
      path1 = pair[0].split(' ')[1]

      if cnt < same_pairs:
         issame = True
      else:
         issame = False
      if os.path.exists(path0) and os.path.exists(path1):  # Only add the pair if both paths exist
         path_list += (path0, path1)
         issame_list.append(issame)
      else:
         print('not exists', path0, path1)
         nrof_skipped_pairs += 1
      cnt += 1
if nrof_skipped_pairs > 0:
      print('Skipped %d image pairs' % nrof_skipped_pairs)

return path_list, issame_list

if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Package  images')
# general
parser.add_argument('--data-dir', default='G:/myFaceData/myvaldata/imgs', help='')
parser.add_argument('--image-size', type=str, default='112,112', help='')
parser.add_argument('--output', default='G:/myFaceData/myvaldata/myval.bin', help='path to save.')
parser.add_argument('--txtfile', default='G:/myFaceData/myvaldata/pairs.txt', help='txtfile path.')
args = parser.parse_args()
image_size = [int(x) for x in args.image_size.split(',')]
img_pairs = read_pairs(args.txtfile)
img_paths, issame_list = get_paths(img_pairs, 5000) # 这里的15925是相同图像对的个数，需要按照实际产生的相同图像对数量替换
img_bins = []
i = 0
for path in img_paths:
      with open(path, 'rb') as fin:
         _bin = fin.read()
         img_bins.append(_bin)
         i += 1
with open(args.output, 'wb') as f:
      pickle.dump((img_bins, issame_list), f, protocol=pickle.HIGHEST_PROTOCOL)

```

欢迎光临 AIHIA梦工厂 (https://aihiamgc.com/)