Question

我正在使用AWS EC2实例上的4个GPU训练Resnet模型。该模型是使用Keras构建的，看起来像这样：

from PIL import Image
import numpy as np
from keras.applications.resnet50 import ResNet50
from keras.layers import Flatten, Dense
from keras.utils import multi_gpu_model
from keras.callbacks import ModelCheckpoint, TensorBoard
from keras import Model

def construct_model():
  base = ResNet50(weights="imagenet", include_top=False, input_shape=(448, 448, 3))
  x = Flatten()(base.output)
  x = Dense(3)(x)
  model = Model(inputs=base.inputs, outputs=x)
  return model

def try_multi_gpu(model):
  try:
    model = multi_gpu_model(model, gpus=4,  cpu_relocation=True)
    print ("Using multi gpu model")
  except ValueError:
    print ("Using single GPU/CPU model")

  return model

def train():
  X_train, Y_train = load_dataset()
  model = construct_model()
  model = try_multi_gpu(model)

  # Set checkpoints
  filepath="weights-improvement-{epoch:02d}-{val_loss:.2f}.hdf5"
  checkpoint = ModelCheckpoint(filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='min')
  callbacks_list = [checkpoint]

  # Create a TensorBoard instance with the path to the logs directory
  tensorboard = TensorBoard(log_dir='logs/{}'.format(time()))
  callbacks_list.append(tensorboard)

  model.fit(X_train, Y_train, initial_epoch=epoch_start ,epochs = 50, batch_size = 10, validation_split=0.15, verbose=1, callbacks=callbacks_list)      model.compile(optimizer='adam', loss='mean_squared_error', metrics=['mae'])

训练完成后，我想通过加载训练后的权重来在本地CPU上运行预测：

def test(imgPath, weights):
  im = Image.open(imgPath)
  w, h = im.size
  im = im.crop((0, (h-w)/2 ,w , (h+w)/2))
  im = im.resize((448, 448), Image.ANTIALIAS)
  imArr = np.array([np.asarray(im)])

  model = construct_model()
  model = try_multi_gpu(model)
  model.load_weights(weights, by_name=True)
  preds = model.predict(imArr)
  print ("Prediction: ", preds)

但是当我尝试运行它时，我遇到了ValueError：

 File "train.py", line 133, in <module>
    test("test/IMG_5767_dark.jpg", "weights-improvement-48-0.01-phi-thetha.hdf5")
  File "train.py", line 115, in test
    model.load_weights(weights, by_name=True)
  File "/usr/local/lib/python3.7/site-packages/keras/engine/saving.py", line 492, in load_wrapper
    return load_function(*args, **kwargs)
  File "/usr/local/lib/python3.7/site-packages/keras/engine/network.py", line 1227, in load_weights
    reshape=reshape)
  File "/usr/local/lib/python3.7/site-packages/keras/engine/saving.py", line 1309, in load_weights_from_hdf5_group_by_name
    ' element(s).')
ValueError: Layer #6 (named "dense_1") expects 2 weight(s), but the saved weights have 0 element(s).

当我在AWS实例上运行相同的预测代码时，它可以工作。有什么方法可以在本地计算机的CPU上运行它吗？

在使用Keras训练了多个GPU之后，是否可以在CPU上进行推理？

0 个答案: