试图编写一个暹罗神经网络。代码到达时代1，然后在那里退出。没有显示错误或异常。
我正在使用fit_generator尝试根据一些参数随机配对图像。我要尝试训练一个暹罗网络来输出特征向量，如果图像匹配，则两个向量之间的距离会很短。
我在Windows x64上。我用pip install tensorflow-gpu安装并安装了cuda 9.0和cudnn 7.2。当我从中获得输出时，Tensorflow肯定会启动，但不确定为什么会失败。
train.py
import os
import cv2
import numpy as np
import pandas as pd
from tqdm import tqdm
import numpy.random as rng
from whalegenerator import WhaleGenerator
import tensorflow as tf
from keras import backend as K
from keras.regularizers import l2
from keras.optimizers import SGD,Adam
from keras.losses import binary_crossentropy
from keras.models import Model, Sequential, save_model, load_model
from keras.callbacks import ModelCheckpoint, EarlyStopping
from keras.layers import Input, Conv2D, Lambda, Subtract, Dense, Flatten,MaxPooling2D
ProcessImages = True
TrainPath = '.\\processed\\train'
ProcessedPath = '.\\processed'
TrainTruthPath = '.\\train.csv'
def buildModel():
#We are building a saimese network
input_shape = (256,256,1)
left = Input(input_shape)
right = Input(input_shape)
convnet = Sequential()
convnet.add(Conv2D(64,(9,9),activation='relu',input_shape=input_shape,kernel_initializer='random_normal',kernel_regularizer=l2(2E-4)))
convnet.add(MaxPooling2D())
convnet.add(Conv2D(128,(7,7),activation='relu',kernel_regularizer=l2(2E-4),kernel_initializer='random_normal',bias_initializer='random_normal'))
convnet.add(MaxPooling2D())
convnet.add(Conv2D(256,(5,5),activation='relu',kernel_initializer='random_normal',kernel_regularizer=l2(2e-4),bias_initializer='random_normal'))
convnet.add(MaxPooling2D())
convnet.add(Conv2D(64,(3,3),activation='relu',kernel_initializer='random_normal',kernel_regularizer=l2(2e-4),bias_initializer='random_normal'))
convnet.add(Flatten())
convnet.add(Dense(2048,activation="sigmoid",kernel_regularizer=l2(1e-3),kernel_initializer='random_normal',bias_initializer='random_normal'))
encodedL = convnet(left)
encodedR = convnet(right)
subtract = Subtract()([encodedL,encodedR])
diff = Lambda(lambda x: K.abs(x))(subtract)
prediction = Dense(1,activation='sigmoid',bias_initializer='random_normal')(diff)
siamese_net = Model(inputs=[left,right],output=prediction)
run_opts = tf.RunOptions(report_tensor_allocations_upon_oom = True)
siamese_net.compile(loss="binary_crossentropy",optimizer=Adam(6E-5),options=run_opts,metrics=['accuracy'])
return siamese_net
model = buildModel()
print(model.count_params())
batch_size = 32
generator = Generator(TrainPath,TrainTruthPath,batch_size,(256,256))
try:
result = model.fit_generator(generator=generator,epochs=100,steps_per_epoch=int(np.floor(len(generator)/batch_size)),max_queue_size=50,verbose=2,callbacks=[
ModelCheckpoint('.\\models\\image_256.h5',save_best_only=True,monitor='accuracy')
])
except ValueError as e:
print(e)
except Exception as e:
print(e)
else:
print('unkown error')
print('finished!')
发电机
import os
import cv2
import random
import numpy as np
import pandas as pd
from keras.utils import Sequence
class Generator(Sequence):
#Number of items in the group
def __len__(self):
return int(np.floor(len(self.imageList)/2))
def __getitem__(self,index):
X = [np.empty((self.batch_size, *self.dim, 1)) for i in range(2)]
Y = np.empty((self.batch_size), dtype=float)
# Generate data
matches = 0
matchKeys = list(self.hasMatches.items())
for i in range(self.batch_size):
if self.batch_size - i < self.min_match and matches < self.min_match:
id1 = random.choice(matchKeys)[0]
images1 = self.imgGroups[id1]
X[0][i,] = images1[random.randint(0,len(images1)-1)]
X[1][i,] = images1[random.randint(0,len(images1)-1)]
Y[i] = 1.0
matches += 1
else:
item1 = self.imageList[random.randint(0,len(self.imageList)-1)]
id1 = item1["Id"]
X[0][i,] = item1["Image"]
images1 = self.imgGroups[id1]
if id1 != "new_whale" and len(images1) > 1 and random.random() <= 0.5:
X[1][i,] = images1[random.randint(0,len(images1)-1)]
Y[i] = 1.0
matches += 1.0
else:
item2 = self.imageList[random.randint(0,len(self.imageList)-1)]
id2 = item2["Id"]
X[1][i,] = item2["Image"]
if id1 != "new_whale" and id1 == id2:
Y[i] = 1.0
matches += 1.0
else:
Y[i] = 0.0
return X, Y
def __init__(self, image_path, csv, batch_size, dim):
self.min_match = 3
self.imgGroups = { }
self.imageList = [ ]
self.hasMatches = { }
self.dim = dim
self.df = pd.read_csv(csv)
self.batch_size = batch_size
for i, row in self.df.iterrows():
_id = row["Id"]
img = cv2.imread(os.path.join(image_path,row["Image"]))
if img is None:
continue
img = img[:,:,0]
img = img.reshape((*self.dim,1))
self.imageList.append({ "Id": _id, "Image": img })
if not _id in self.imgGroups:
self.imgGroups[_id] = [img]
else:
if _id != "new_whale":
self.hasMatches[_id] = True
self.imgGroups[_id].append(img)
输出
Using TensorFlow backend.
83297857
Epoch 1/100
2018-12-21 21:31:59.191422: I T:\src\github\tensorflow\tensorflow\core\platform\cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2018-12-21 21:32:00.023587: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1405] Found device 0 with properties:
name: GeForce GTX 980M major: 5 minor: 2 memoryClockRate(GHz): 1.1265
pciBusID: 0000:01:00.0
totalMemory: 4.00GiB freeMemory: 3.32GiB
2018-12-21 21:32:00.035814: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1484] Adding visible gpu devices: 0
2018-12-21 21:32:00.460105: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:965] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-12-21 21:32:00.467842: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:971] 0
2018-12-21 21:32:00.473743: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:984] 0: N
2018-12-21 21:32:00.478649: I T:\src\github\tensorflow\tensorflow\core\common_runtime\gpu\gpu_device.cc:1097] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3039 MB memory) -> physical GPU (device: 0, name: GeForce GTX 980M, pci bus id: 0000:01:00.0, compute capability: 5.2)
答案 0 :(得分：0)
默认情况下，创建TensorFlow以使用特定类型的AVX（高级矢量指令）指令。您的CPU支持输出中提到的AVX2，但TensorFlow并未使用它。