前往小程序,Get更优阅读体验!
立即前往
首页
学习
活动
专区
工具
TVP
发布
社区首页 >专栏 >人脸检测——AFLW准备人脸

人脸检测——AFLW准备人脸

作者头像
MachineLP
发布2018-01-09 14:53:18
2.6K0
发布2018-01-09 14:53:18
举报
文章被收录于专栏:小鹏的专栏小鹏的专栏

不多说了,直接代码吧:

生成AFLW_ann.txt的代码,其中包含图像名称 和 图像中人脸的位置(x,y,w,h);

** AFLW中含有aflw.aqlite文件。

代码语言:javascript
复制
import sqlite3

list_annotation = list()
# Format for saving: path x y w h
ann_format = "{}/{} {} {} {} {}"

conn = sqlite3.connect('aflw.sqlite')

fidQuery = 'SELECT face_id FROM Faces'
faceIDs = conn.execute(fidQuery)

for idx in faceIDs:

    fidQuery = 'SELECT file_id FROM Faces WHERE face_id = {}'.format(idx[0])
    imgID = conn.execute(fidQuery)
    imgID = [id for id in imgID]

    imgDataQuery = "SELECT db_id,filepath,width,height FROM FaceImages WHERE file_id = '{}'".format(imgID[0][0])
    fileID = conn.execute(imgDataQuery)
    fileID = [id for id in fileID]
    db_id = fileID[0][0]
    filepath = fileID[0][1]

    faceRectQuery = 'SELECT x,y,w,h FROM FaceRect WHERE face_id = {}'.format(idx[0])
    faceRect = conn.execute(faceRectQuery)
    faceRect = [id for id in faceRect]

    if len(faceRect)==0:
        continue
    
    x,y,w,h =  faceRect[0]

    list_annotation.append(ann_format.format(db_id,filepath,x,y,w,h))


with open("AFLW_ann.txt",'w') as f:
    f.writelines("%s\n" % line for line in list_annotation) 

AFLW图片都整理到flickr文件下(含0,1,2三个文件),生成人脸的程序(并且对人脸进行了左右镜像):

代码语言:javascript
复制
import os
from PIL import Image
from PIL import ImageFile
# ImageFile.LOAD_TRUNCATED_IMAGES = True
import cv2
import numpy as np

with open('AFLW_ann.txt','r') as f:
    lines = f.readlines()

save_dir1 = 'data_prepare/net_positive'
save_dir2 = 'data_prepare/net_positive_flip'

if os.path.exists(save_dir1)==False:
    os.makedirs(save_dir1)
if os.path.exists(save_dir2)==False:
    os.makedirs(save_dir2)

for idx, line in enumerate(lines):
    s1 = line.strip().split(' ')
    image_path = s1[0]
    x = int(s1[1])
    y = int(s1[2])
    w = int(s1[3])
    h = int(s1[4])
    print (image_path)
    # image = Image.open(image_path)
    image = cv2.imread(image_path)
    if image is None:
        continue
    if x<=0 and y<=0 and w<=0 and h<=0:
            continue
    box = (x, y, x+w, y+h)
    
    # patch = image.crop(box)
    patch = image[box[1]:box[3], box[0]:box[2], :]
    if patch is None:
        continue

    patch1 = patch #.resize((51, 51))
    # patch2 = patch1.transpose(Image.FLIP_LEFT_RIGHT)
    h = patch.shape[0]
    w = patch.shape[1] 
    iLR = patch.copy()
    for i in range(h):
       for j in range(w):
           iLR[i,w-1-j] = patch[i,j]
    patch2 = iLR

    s2 = image_path.split('/')
    image_name = s2[-1]

    save_path1 = save_dir1+'/'+str(idx)+image_name + '.jpg'
    save_path2 = save_dir2+'/'+'l'+str(idx)+image_name + '.jpg'

    #patch1.save(save_path1, 'jpeg')
    #patch2.save(save_path2, 'jpeg')
    cv2.imwrite(save_path1, np.array(patch1))
    cv2.imwrite(save_path2, np.array(patch2))

    print (idx)
本文参与 腾讯云自媒体分享计划,分享自作者个人站点/博客。
原始发表:2017年07月07日,如有侵权请联系 cloudcommunity@tencent.com 删除

本文分享自 作者个人站点/博客 前往查看

如有侵权,请联系 cloudcommunity@tencent.com 删除。

本文参与 腾讯云自媒体分享计划  ,欢迎热爱写作的你一起参与!

评论
登录后参与评论
0 条评论
热度
最新
推荐阅读
相关产品与服务
图像处理
图像处理基于腾讯云深度学习等人工智能技术,提供综合性的图像优化处理服务,包括图像质量评估、图像清晰度增强、图像智能裁剪等。
领券
问题归档专栏文章快讯文章归档关键词归档开发者手册归档开发者手册 Section 归档