Inception-v 3既成重みによる特徴抽出(画像認識)


tensorflow公式サイトの画像認識の中国語紹介では、Tensorflowのモデルコードライブラリのclassify_image.pyで画像認識を行います.どのようにテストするか、そして最後の層の1*1*2048次元の特徴抽出方法も紹介されているので、ここで紹介します.
......

with tf.Session() as sess:
    # Some useful tensors:
    # 'softmax:0': A tensor containing the normalized prediction across
    #   1000 labels.
    # 'pool_3:0': A tensor containing the next-to-last layer containing 2048
    #   float description of the image.
    # 'DecodeJpeg/contents:0': A tensor containing a string providing JPEG
    #   encoding of the image.
    # Runs the softmax tensor by feeding the image_data as input to the graph.
    softmax_tensor = sess.graph.get_tensor_by_name('softmax:0')

.......

ソースコードはここで紹介したもので、3つのインタフェースがあり、
'softmax:0': A tensor containing the normalized prediction across 1000 labels.
'pool_3:0': A tensor containing the next-to-last layer containing 2048 float description of the image.
'DecodeJpeg/contents:0': A tensor containing a string providing JPEG encoding of the image.
予測すると直接'softmax:0':'DecodeJpeg/contents:0':画像認識のテストが可能
フィーチャーを抽出するには
            fc_tensor = sess.graph.get_tensor_by_name('pool_3:0')
            pool_1 = sess.run(fc_tensor,{'DecodeJpeg/contents:0': image_data})

はい、保存するならCSVか選択できます.matファイル
import tensorflow as tf
import numpy as np
import os
from PIL import Image
import matplotlib.pyplot as plt
import scipy.io as scio

model_dir='F:/fqh/models-master/tutorials/image/imagenet/2015'
image = 'F:/fqh/models-master/tutorials/image/imagenet/data_set/face/faces95_72_20_180-200jpgfar-close/'

target_path=image+'wjhugh/'
class NodeLookup(object):
    def __init__(self, label_lookup_path=None, uid_lookup_path=None):
        if not label_lookup_path:
            label_lookup_path = os.path.join(
                    model_dir, 'imagenet_2012_challenge_label_map_proto.pbtxt')
        if not uid_lookup_path:
            uid_lookup_path = os.path.join(
                    model_dir, 'imagenet_synset_to_human_label_map.txt')
        self.node_lookup = self.load(label_lookup_path, uid_lookup_path)

    def load(self, label_lookup_path, uid_lookup_path):
        if not tf.gfile.Exists(uid_lookup_path):
            tf.logging.fatal('File does not exist %s', uid_lookup_path)
        if not tf.gfile.Exists(label_lookup_path):
            tf.logging.fatal('File does not exist %s', label_lookup_path)

        proto_as_ascii_lines = tf.gfile.GFile(uid_lookup_path).readlines()
        
        uid_to_human = {}
        for line in proto_as_ascii_lines:

            line = line.strip('
') parse_items = line.split('\t') uid = parse_items[0] human_string = parse_items[1] uid_to_human[uid] = human_string proto_as_ascii = tf.gfile.GFile(label_lookup_path).readlines() node_id_to_uid = {} for line in proto_as_ascii: if line.startswith(' target_class:'): target_class = int(line.split(': ')[1]) if line.startswith(' target_class_string:'): target_class_string = line.split(': ')[1] node_id_to_uid[target_class] = target_class_string[1:-2] node_id_to_name = {} for key, val in node_id_to_uid.items(): if val not in uid_to_human: tf.logging.fatal('Failed to locate: %s', val) name = uid_to_human[val] node_id_to_name[key] = name return node_id_to_name def id_to_string(self, node_id): if node_id not in self.node_lookup: return '' return self.node_lookup[node_id] def create_graph(): with tf.gfile.FastGFile(os.path.join( model_dir, 'classify_image_graph_def.pb'), 'rb') as f: graph_def = tf.GraphDef() graph_def.ParseFromString(f.read()) tf.import_graph_def(graph_def, name='') create_graph() with tf.Session() as sess: softmax_tensor = sess.graph.get_tensor_by_name('softmax:0') for root, dirs, files in os.walk(target_path): for file in files: # print(file) img_path = target_path+file image_data = tf.gfile.FastGFile(img_path, 'rb').read() fc_tensor = sess.graph.get_tensor_by_name('pool_3:0') pool_1 = sess.run(fc_tensor,{'DecodeJpeg/contents:0': image_data}) # print(pool_1) img_path=img_path[:len(img_path)-4] #print(img_path) scio.savemat(img_path+'.mat', {"pool_1": pool_1})

兄弟子は自分のデータセットの画像の特徴を抽出する必要があるので、このように書きました.もう一つのサイクルを加えて、データセット全体を遍歴することもできます.コンピュータの配置が限られているので、このように書きました.私が変更したソース+重み
新しい変更
import tensorflow as tf
import numpy as np
import os
from PIL import Image
import matplotlib.pyplot as plt
import scipy.io as scio

model_dir='F:/fqh/models-master/tutorials/image/imagenet/2015'
image = 'F:/fqh/models-master/tutorials/image/imagenet/data_set/face/faces96_152_20_180-200jpgview-depth/'

target_path=image+'wjhugh/'
class NodeLookup(object):
    def __init__(self, label_lookup_path=None, uid_lookup_path=None):
        if not label_lookup_path:
            label_lookup_path = os.path.join(
                    model_dir, 'imagenet_2012_challenge_label_map_proto.pbtxt')
        if not uid_lookup_path:
            uid_lookup_path = os.path.join(
                    model_dir, 'imagenet_synset_to_human_label_map.txt')
        self.node_lookup = self.load(label_lookup_path, uid_lookup_path)

    def load(self, label_lookup_path, uid_lookup_path):
        if not tf.gfile.Exists(uid_lookup_path):
            tf.logging.fatal('File does not exist %s', uid_lookup_path)
        if not tf.gfile.Exists(label_lookup_path):
            tf.logging.fatal('File does not exist %s', label_lookup_path)

        proto_as_ascii_lines = tf.gfile.GFile(uid_lookup_path).readlines()
        
        uid_to_human = {}
        for line in proto_as_ascii_lines:

            line = line.strip('
') parse_items = line.split('\t') uid = parse_items[0] human_string = parse_items[1] uid_to_human[uid] = human_string proto_as_ascii = tf.gfile.GFile(label_lookup_path).readlines() node_id_to_uid = {} for line in proto_as_ascii: if line.startswith(' target_class:'): target_class = int(line.split(': ')[1]) if line.startswith(' target_class_string:'): target_class_string = line.split(': ')[1] node_id_to_uid[target_class] = target_class_string[1:-2] node_id_to_name = {} for key, val in node_id_to_uid.items(): if val not in uid_to_human: tf.logging.fatal('Failed to locate: %s', val) name = uid_to_human[val] node_id_to_name[key] = name return node_id_to_name def id_to_string(self, node_id): if node_id not in self.node_lookup: return '' return self.node_lookup[node_id] def create_graph(): with tf.gfile.FastGFile(os.path.join( model_dir, 'classify_image_graph_def.pb'), 'rb') as f: graph_def = tf.GraphDef() graph_def.ParseFromString(f.read()) tf.import_graph_def(graph_def, name='') create_graph() list0=[] for root, dirs,files in os.walk(image): list0.append(dirs) #print(list0[0]) img_list=[] # print(img_list) for ii in list0[0]: img_list.append(ii) list_img_name=np.array(img_list) list_img_name.sort() # print(list_img_name[0]) with tf.Session() as sess: softmax_tensor = sess.graph.get_tensor_by_name('softmax:0') for jj in range(0,len(list_img_name)):#len(list_img_name) target_path=image+list_img_name[jj]+'/' for root, dirs, files in os.walk(target_path): for file in files: img_path = target_path+file image_data = tf.gfile.FastGFile(img_path, 'rb').read() fc_tensor = sess.graph.get_tensor_by_name('pool_3:0') pool_1 = sess.run(fc_tensor,{'DecodeJpeg/contents:0': image_data}) pool_2 = pool_1[0,0,0,:] img_path=img_path[:len(img_path)-4] scio.savemat(img_path+'.mat', {"pool_2": pool_2}) pi= (jj/(len(list_img_name)-1))*100 print("%4.2f %%" % pi)

ベクトルをまっすぐにしてデータセット全体を巡回