insight face数据打包（一）：LFW

LFW数据集->rec文件

• 数据集准备：
  • raw data下载：(提供两种来源）
    • http://vis-www.cs.umass.edu/lfw/index.html#download 
  • 已经对齐好到bin文件：https://download.csdn.net/download/ustczhng2012/15002984
• 数据对齐
  • 基于原图+landmark点对齐图片
  • LFW直接下载的raw data没有landmark点，一般需要landmark模型生成关键点，然后旋转缩放对齐后再生成bin文件

import cv2
import numpy as np
from skimage import transform as trans

src1 = np.array([[51.642, 50.115], [57.617, 49.990], [35.740, 69.007],
                 [51.157, 89.050], [57.025, 89.702]],
                dtype=np.float32)
#<--left
src2 = np.array([[45.031, 50.118], [65.568, 50.872], [39.677, 68.111],
                 [45.177, 86.190], [64.246, 86.758]],
                dtype=np.float32)

#---frontal
src3 = np.array([[39.730, 51.138], [72.270, 51.138], [56.000, 68.493],
                 [42.463, 87.010], [69.537, 87.010]],
                dtype=np.float32)

#-->right
src4 = np.array([[46.845, 50.872], [67.382, 50.118], [72.737, 68.111],
                 [48.167, 86.758], [67.236, 86.190]],
                dtype=np.float32)

#-->right profile
src5 = np.array([[54.796, 49.990], [60.771, 50.115], [76.673, 69.007],
                 [55.388, 89.702], [61.257, 89.050]],
                dtype=np.float32)

src = np.array([src1, src2, src3, src4, src5])
src_map = {112: src, 224: src * 2}

arcface_src = np.array(
    [[38.2946, 51.6963], [73.5318, 51.5014], [56.0252, 71.7366],
     [41.5493, 92.3655], [70.7299, 92.2041]],
    dtype=np.float32)

arcface_src = np.expand_dims(arcface_src, axis=0)

# lmk is prediction; src is template
def estimate_norm(lmk, image_size=112, mode='arcface'):
    assert lmk.shape == (5, 2)
    tform = trans.SimilarityTransform()
    lmk_tran = np.insert(lmk, 2, values=np.ones(5), axis=1)
    min_M = []
    min_index = []
    min_error = float('inf')
    if mode == 'arcface':
        assert image_size == 112
        src = arcface_src
    else:
        src = src_map[image_size]
    for i in np.arange(src.shape[0]):
        tform.estimate(lmk, src[i])
        M = tform.params[0:2, :]
        results = np.dot(M, lmk_tran.T)
        results = results.T
        error = np.sum(np.sqrt(np.sum((results - src[i])**2, axis=1)))
        #         print(error)
        if error < min_error:
            min_error = error
            min_M = M
            min_index = i
    return min_M, min_index


def norm_crop(img, landmark, image_size=112, mode='arcface'):
    M, pose_index = estimate_norm(landmark, image_size, mode)
    warped = cv2.warpAffine(img, M, (image_size, image_size), borderValue=0.0)
    return warped

• 生成rec和idx文件
  • 可以参考emore.rec生成方法，使用该方法不需要上述数据对齐脚本，里面自带对齐脚本，只需要配置好图片配置txt文件即可
• 创建pair文件
  • 创建pairs.txt脚本：（参考链接）

import glob
import os.path
import numpy as np
import os

def find_not_zero_pos(sstr):
    for i in range(len(sstr)):
        if sstr[i] != "0":
            return i
    return len(sstr) - 1

def get_real_str(sstr):
    i = find_not_zero_pos(sstr)
    return sstr[i:]

def create_match_content():
    matched_result = set()
    k = 0
    sub_dirs = [x[0] for x in os.walk(INPUT_DATA)]
    while len(matched_result) < 3000:
        for sub_dir in sub_dirs[1:]:
            extensions = 'jpg'
            file_list = []
            dir_name = os.path.basename(sub_dir)
            file_glob = os.path.join(INPUT_DATA, dir_name, '*.' + extensions)
            # glob.glob(file_glob)获取指定目录下的所有图片
            file_list.extend(glob.glob(file_glob))
            if not file_list:
                continue
            if len(file_list) >= 2:
                label_name = dir_name
                length = len(file_list)
                random_number1 = np.random.randint(length)
                random_number2 = np.random.randint(length)
                while random_number1 == random_number2:
                    random_number1 = np.random.randint(length)
                    random_number2 = np.random.randint(length)
                base_name1 = os.path.basename(file_list[random_number1 % length])
                base_name2 = os.path.basename(file_list[random_number2 % length])
                if (file_list[random_number1 % length] != file_list[random_number2 % length]):
                    base_name1 = base_name1.split("_")[-1]
                    base_name1 = base_name1.split(".")[0]
                    base_name2 = base_name2.split("_")[-1]
                    base_name2 = base_name2.split(".")[0]
                    matched_result.add(label_name + ' ' + get_real_str(base_name1) + ' ' + get_real_str(base_name2))
                    # print(label_name + ' ' + get_real_str(base_name1) + ' ' + get_real_str(base_name2))
                    k = k + 1
    return matched_result, k

def create_unmatch_content():
    """不同类的匹配对"""
    unmatched_result = set()
    k = 0
    while len(unmatched_result) < 3000:
        sub_dirs = [x[0] for x in os.walk(INPUT_DATA)]
        length_of_dir = len(sub_dirs)
        for j in range(24):
            for i in range(1, length_of_dir):
                class1 = sub_dirs[i]
                random_number = np.random.randint(length_of_dir)
                while random_number == 0 | random_number == i:
                    random_number = np.random.randint(length_of_dir)
                class2 = sub_dirs[random_number]
                class1_name = os.path.basename(class1)
                class2_name = os.path.basename(class2)
                extensions = 'jpg'
                file_list1 = []
                file_list2 = []
                file_glob1 = os.path.join(INPUT_DATA, class1_name, '*.' + extensions)
                file_list1.extend(glob.glob(file_glob1))
                file_glob2 = os.path.join(INPUT_DATA, class2_name, '*.' + extensions)
                file_list2.extend(glob.glob(file_glob2))
                if file_list1 and file_list2:
                    base_name1 = os.path.basename(file_list1[j % len(file_list1)])
                    base_name2 = os.path.basename(file_list2[j % len(file_list2)])
                    base_name1 = base_name1.split("_")[-1]
                    base_name1 = base_name1.split(".")[0]
                    base_name1 = get_real_str(base_name1)
                    base_name2 = base_name2.split("_")[-1]
                    base_name2 = base_name2.split(".")[0]
                    base_name2 = get_real_str(base_name2)
                    s = class2_name + ' ' + base_name2 + ' ' + class1_name + ' ' + base_name1
                    if (s not in unmatched_result):
                        unmatched_result.add(s)
                        if len(unmatched_result) > 3000:
                            break
                    k = k + 1
    return unmatched_result, k

if __name__ == '__main__':
    INPUT_DATA = r'dataset/val/lfw-deepfunneled'
    txt_path = 'dataset/val/pairs.txt'
    if os.path.isfile(txt_path):
        os.remove(txt_path)
    result, k1 = create_match_content()
    print(k1)
    # print(result)
    result_un, k2 = create_unmatch_content()
    print(k2)
    # print(result_un)
    file = open(txt_path, 'w')
    result1 = list(result)
    result2 = list(result_un)
    file.write('10 300\n')
    for i in range(10):
        for pair in result1[i * 300:i * 300 + 300]:
            file.write(pair + '\n')
        for pair in result2[i * 300:i * 300 + 300]:
            file.write(pair + '\n')
    file.close()

• 生成验证集bin文件
  • 注意：生成到bin文件数据是BGR格式的，在使用时需要注意

import pickle
import os
from tqdm import tqdm
import cv2
if __name__ == '__main__':
    lfw_dir = r'dataset/val/lfw-deepfunneled'
    image_size = (112, 112)
    pairs_path = r'dataset/val/pairs.txt'
    bin_output = r'dataset/val/lfw.bin'
    lfw_bins = []
    issame_list = []
    with open(pairs_path, 'r') as f:
        lines = f.readlines()
        line = lines[0].strip().split(' ')
        for i in tqdm(range(1, len(lines))):
            temp = lines[i].strip().split(' ')
            if len(temp) == 3:
                img_path_1 = os.path.join(lfw_dir, temp[0], temp[0]+'_'+temp[1].zfill(4)+'.jpg')
                img_1 = cv2.imread(img_path_1)
                img_1 = cv2.resize(img_1, image_size)
                img_save_path_1 = os.path.join(lfw_dir, temp[0], temp[0] + '_' + temp[1].zfill(4) + '_resize.jpg')
                cv2.imwrite(img_save_path_1, img_1)
                with open(img_save_path_1, 'rb') as fin:
                    _bin_1 = fin.read()
                    lfw_bins.append(_bin_1)

                img_path_2 = os.path.join(lfw_dir, temp[0], temp[0]+'_'+temp[2].zfill(4)+'.jpg')
                img_2 = cv2.imread(img_path_2)
                img_2 = cv2.resize(img_2, image_size)
                img_save_path_2 = os.path.join(lfw_dir, temp[0], temp[0]+'_'+temp[2].zfill(4)+'_resize.jpg')
                cv2.imwrite(img_save_path_2, img_2)
                with open(img_save_path_2, 'rb') as fin:
                    _bin_2 = fin.read()
                    lfw_bins.append(_bin_2)
                issame_list.append(1)
            elif len(temp) == 4:
                img_path_1 = os.path.join(lfw_dir, temp[0], temp[0]+'_'+temp[1].zfill(4)+'.jpg')
                img_1 = cv2.imread(img_path_1)
                img_1 = cv2.resize(img_1, image_size)
                img_save_path_1 = os.path.join(lfw_dir, temp[0], temp[0] + '_' + temp[1].zfill(4) + '_resize.jpg')
                cv2.imwrite(img_save_path_1, img_1)
                with open(img_save_path_1, 'rb') as fin:
                    _bin_1 = fin.read()
                    lfw_bins.append(_bin_1)

                img_path_2 = os.path.join(lfw_dir, temp[2], temp[2]+'_'+temp[3].zfill(4)+'.jpg')
                img_2 = cv2.imread(img_path_2)
                img_2 = cv2.resize(img_2, image_size)
                img_save_path_2 = os.path.join(lfw_dir, temp[2], temp[2] + '_' + temp[3].zfill(4) + '_resize.jpg')
                cv2.imwrite(img_save_path_2, img_2)
                with open(img_save_path_2, 'rb') as fin:
                    _bin_2 = fin.read()
                    lfw_bins.append(_bin_2)
                issame_list.append(0)
            else:
                print("drop this line: %d" %i)
                continue
    with open(bin_output, 'wb') as f:
        pickle.dump((lfw_bins, issame_list), f, protocol=pickle.HIGHEST_PROTOCOL)