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
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：

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。

#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)