keras model.load_weights 错误 NoneType' object 没有属性 'fit'

Question

hope someone can help with this, I am trying to train a model base on another pre-trained model, but I keep getting this error.

Thanks in advance. and happy coding.

Model

model = Unet(backbone_name=efficientnetb2, encoder_weights='imagenet',
             input_shape=(256, 256, 3),
             classes=1, activation='sigmoid')

    model = get_model(
        mparams['backbone'], 
        input_shape=(mparams['img_size'], mparams['img_size'], 3),
        loss_type=mparams['loss'],
        umodel=mparams['umodel'],
        lr=mparams['lr']
    )
    
    # load model
    model = model.load_weights('./models_v24/model_0.hdf5')

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-19-b3a8ba8f71c6> in <module>
     41             resize=None
     42         )
---> 43         model, history = train_model(PARAMS, n, train_datagen, val_datagen)
     44         plt.plot(history.history['loss'], label='loss')
     45         plt.plot(history.history['val_loss'], label='val_loss')

<ipython-input-17-191deeedc4ed> in train_model(mparams, n_fold, train_datagen, val_datagen)
     53     model = model.load_weights('./models_v24/model_0.hdf5')
     54 
---> 55     history = model.fit(
     56                         train_datagen,
     57                         validation_data=val_datagen,

AttributeError: 'NoneType' object has no attribute 'fit'

Answer 1

Your comment # load model which is not correct. You are just loading weights here, not the whole model . This function returns None when loading weights in HDF5 format which explains the current error here.

If the model is created, you just need model.load_weights('./models_v24/model_0.hdf5') . Then you should be able to fit the model.

model = get_model(...)

# load weights
model.load_weights('./models_v24/model_0.hdf5')

Also I noticed you did not accept some of the answers in SO. Please take a look at What should I do when someone answers my question?

Answer 2

don't use [model = model.load_weights()] use [model.load_weights()] direactly