在 Keras 中保存模型时引发“无法创建组（名称已存在）”错误

Question

我正在使用来自 keras.applications 的 ResNet50 和 DenseNet121 构建模型融合，但在保存模型时引发错误。如果我只使用 ResNet50 和 DenseNet121 的一个网络，例如 DenseNet only，没问题

与 ResNet50 和 DenseNet121 的融合：

img_input = Input(shape=input_shape)

densenet = app.DenseNet121(
    include_top=False,
    input_tensor=img_input,
    input_shape=input_shape,
    weights=base_weights)
resnet = app.ResNet50(
    include_top=False,
    input_tensor=img_input,
    input_shape=input_shape,
    weights=base_weights)

x1 = densenet.output
x1 = GlobalAveragePooling2D(name='dn_gap_last')(x1)
# then x1.shape is (batch, 1024)

x2 = resnet.output
x2 = Flatten()(x2)  # then x2.shape is (batch, 2048)

x = concatenate([x1, x2], axis=-1)
predictions = Dense(len(class_names), activation="sigmoid", name="predictions")(x)
model = Model(inputs=img_input, outputs=predictions)

并通过 ModelCheckpoint 保存模型

checkpoint = ModelCheckpoint(
                 output_weights_path,
                 save_weights_only=True,
                 save_best_only=True,
                 verbose=1,
            )

但在保存 mdoel 时引发错误

Epoch 00001: val_loss improved from inf to 0.72018, saving model to ./experiments/8/weights.h5
Traceback (most recent call last):
  File "train.py", line 229, in <module>
    main()
  File "train.py", line 212, in main
    shuffle=False,
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/legacy/interfaces.py", line 91, in wrapper
    return func(*args, **kwargs)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/engine/training.py", line 2280, in fit_generator
    callbacks.on_epoch_end(epoch, epoch_logs)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/callbacks.py", line 77, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/callbacks.py", line 445, in on_epoch_end
    self.model.save_weights(filepath, overwrite=True)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/engine/topology.py", line 2607, in save_weights
    save_weights_to_hdf5_group(f, self.layers)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/keras/engine/topology.py", line 2878, in save_weights_to_hdf5_group
    g = f.create_group(layer.name)
  File "/home/hqt/chest-x-ray-project/code/venv/lib/python3.6/site-packages/h5py/_hl/group.py", line 50, in create_group
    gid = h5g.create(self.id, name, lcpl=lcpl)
  File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
  File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
  File "h5py/h5g.pyx", line 151, in h5py.h5g.create
ValueError: Unable to create group (name already exists)

Answer 1

如果我需要像您的情况一样使用像 tf.tile 这样的操作，我将使用 lambda 层来调用它。所以有效的代码如下

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Input, Lambda
from tensorflow.keras import Model

def my_fun(a):
  out = tf.tile(a, (1, tf.shape(a)[0]))
  return out

a = Input(shape=(10,))
#out = tf.tile(a, (1, tf.shape(a)[0]))
out = Lambda(lambda x : my_fun(x))(a)
model = Model(a, out)

x = np.zeros((50,10), dtype=np.float32)
print(model(x).numpy())

model.save('my_model.h5')

#load the model
new_model=tf.keras.models.load_model("my_model.h5")

任何遇到类似问题的人，请关注与此问题相关的GitHub 问题以获得最终解决方案。谢谢！

编辑于 02/04/2020

通过最近的代码修改，您可以使用tf-nightly“h5”格式保存模型，而不会出现任何问题，如下所示。

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Input
from tensorflow.keras import Model

a = Input(shape=(10,))
out = tf.tile(a, (1, tf.shape(a)[0]))
model = Model(a, out)

x = np.zeros((50,10), dtype=np.float32)
print(model(x).numpy())

model.save('./my_model', save_format='h5')