Keras_Segmentation VGG-Unet 导致 AttributeError: “Functional”对象没有属性“output_width”；

Question

我使用 image-segmentation-keras 库 创建了一个模型，如下初始化：

import keras_segmentation
from keras_segmentation.models.unet import vgg_unet
from tensorflow.keras.layers import Input

model = vgg_unet(n_classes=21 ,  input_height=256, input_width=448)

然后我这样训练它：

model.train(
    train_images =  "/content/drive/MyDrive/imgs_train/",
    train_annotations = "/content/drive/MyDrive/masks_train/",
    val_images =  "/content/drive/MyDrive/mgs_validation/",
    val_annotations = "/content/drive/MyDrive/masks_validation/",
    checkpoints_path = "/content/drive/MyDrive/tmp/vgg_unet_1" , 
    epochs=28,validate=True,callbacks = [myCallback])

model.load_weights('checkpoint_filepath')

并像这样保存它：

model.save('/content/drive/MyDrive/vgg_unet_segmentation.h5')

然后像这样加载它：

model = load_model('/content/drive/MyDrive/vgg_unet_segmentation.h5'')

但是，当我尝试通过执行 out = model.predict_segmentation(inp=inp, out_fname="/tmp/out.png")，我收到以下错误：

AttributeError: 'Functional' object has no attribute 'predict_segmentation'

因此，为了解决此问题，我执行了以下操作：

from types import MethodType
model = load_model('/content/drive/MyDrive/vgg_unet_segmentation.h5'')
model.predict_segmentation = MethodType(keras_segmentation.predict.predict, model)

但是，这导致了另一个我无法解决的问题：

[<ipython-input-7-a4b7d02cd9a2>](https://localhost:8080/#) in <module>()
      4 out = model.predict_segmentation(
      5     inp=inp,
----> 6     out_fname="/tmp/out.png")

[/content/image-segmentation-keras/keras_segmentation/predict.py](https://localhost:8080/#) in predict(model, inp, out_fname, checkpoints_path, overlay_img, class_names, show_legends, colors, prediction_width, prediction_height, read_image_type)
    148     assert (len(inp.shape) == 3 or len(inp.shape) == 1 or len(inp.shape) == 4), "Image should be h,w,3 "
    149 
--> 150     output_width = model.output_width
    151     output_height = model.output_height
    152     input_width = model.input_width

AttributeError: 'Functional' object has no attribute 'output_width'

任何知道为什么会发生这种情况，如果是这样，如何解决？

任何帮助表示赞赏！

谢谢！

Answer 1

尝试像以下代码那样的输出尝试model.predict（）：

prediction = (model.predict(test_img_input))
predicted_img=np.argmax(prediction, axis=3)[0,:,:]
plt.imshow(predicted_img)