【Caffe实践】基于CNN的性别、年龄识别的代码实现

主要参考：http://www.openu.ac.il/home/hassner/projects/cnn_agegender/

具体实现：http://nbviewer.jupyter.org/url/www.openu.ac.il/home/hassner/projects/cnn_agegender/cnn_age_gender_demo.ipynb

首先按照链接：http://www.openu.ac.il/home/hassner/projects/cnn_agegender/下载所需的cnn_age_gender_models_and_data.0.0.2.zip到models。

先要保证python版本2.7以上http://yijiebuyi.com/blog/108ae6186bb00cc708bc54f02adec277.html

升级python后，可能导致pip不可用，执行以下命令即可

wget https://bootstrap.pypa.io/get-pip.py

python get-pip.py

numpy路径：/usr/lib64/python2.7/site-packages/numpy/core/include/numpy/arrayobject.h

之后需要编译caffe，使其支持python，具体编译方法见：http://caffe.berkeleyvision.org/installation.html。

预测代码

[python] view plain copy

1. import os  
2. import numpy as np  
3. import shutil  
4.   
5. caffe_root = '/opt/caffe/'   
6. import sys  
7. sys.path.insert(0, caffe_root + 'python')  
8. import caffe  
9.   
10. def predict(src_folder):  
11.     mean_filename='/opt/caffe/models/AgeGenderCNN/mean.binaryproto'  
12.     proto_data = open(mean_filename, "rb").read()  
13.     a = caffe.io.caffe_pb2.BlobProto.FromString(proto_data)  
14.     mean  = caffe.io.blobproto_to_array(a)[0]  
15.   
16.     age_net_pretrained='/opt/caffe/models/AgeGenderCNN/age_net.caffemodel'  
17.     age_net_model_file='/opt/caffe/models/AgeGenderCNN/deploy_age.prototxt'  
18.     age_net = caffe.Classifier(age_net_model_file, age_net_pretrained,  
19.                            mean=mean,  
20.                            channel_swap=(2,1,0),  
21.                            raw_scale=255,  
22.                            image_dims=(256, 256))  
23.   
24.     gender_net_pretrained='/opt/caffe/models/AgeGenderCNN/gender_net.caffemodel'  
25.     gender_net_model_file='/opt/caffe/models/AgeGenderCNN/deploy_gender.prototxt'  
26.     gender_net = caffe.Classifier(gender_net_model_file, gender_net_pretrained,  
27.                            mean=mean,  
28.                            channel_swap=(2,1,0),  
29.                            raw_scale=255,  
30.                            image_dims=(256, 256))  
31.   
32.     age_list=['(0, 2)','(4, 6)','(8, 12)','(15, 20)','(25, 32)','(38, 43)','(48, 53)','(60, 100)']  
33.     gender_list=['Male','Female']  
34.     gender_folder = 'male'  
35.     for people_folder in os.listdir(src_folder):  
36.         people_path = src_folder + people_folder + '/'  
37.         for img_file in os.listdir(people_path):  
38.             img_path = people_path+img_file  
39.             input_image = caffe.io.load_image(img_path)  
40. #            prediction = age_net.predict([input_image])   
41. #            print 'predicted age:', age_list[prediction[0].argmax()]  
42.             prediction = gender_net.predict([input_image])  
43. #            print 'predicted gender:', gender_list[prediction[0].argmax()]  
44.             if gender_list[prediction[0].argmax()] != gender_folder:  
45.                 print 'processing img:',img_path ,'gender:',gender_list[prediction[0].argmax()],' prediction:',prediction  
46.                 if gender_folder == 'Male':  
47.                     shutil.copy(img_path,src_folder+'../maleout')  
48.                 elif gender_folder == 'Female':  
49.                     shutil.copy(img_path,src_folder+'../femaleout')  
50.   
51. if __name__ == '__main__':  
52.     if len(sys.argv) != 2:  
53.         print 'Usage: python %s src_folder' % (sys.argv[0])  
54.         sys.exit()  
55.     src_folder = sys.argv[1]  
56.     if not src_folder.endswith('/'):  
57.         src_folder += '/'  
58.     predict(src_folder)