mmxmb / keras_ensemblng Goto Github PK

@LawnboyMax
hi, i met this situation.

RuntimeError: Graph disconnected: cannot obtain value for tensor Tensor("input_1_3:0", shape=(?, 224, 224, 3), dtype=float32) at layer "input_1". The following previous layers were accessed without issue: []

i check the code,and it seem that the Model in the ensemble function
how to change it ?

here is my code:

model_input= Input(shape=(224,224,3))

model_vgg = model_from_json(open('model_472.json').read())
model_vgg.load_weights('weights_472.h5')

model_inc = model_from_json(open('model_473.json').read()) model_inc.load_weights('weights_473.h5')

model_dense = model_from_json(open('model_47.json').read())
model_dense.load_weights('weights_47.h5')
models = [model_vgg,model_inc,model_dense]

when i call the ensemble function:

def ensemble(models,model_input):
outputs = [model.outputs[0] for model in models]
y = Average()(outputs)
model = Model(model_input,y,name='ensemble')
return mode
ensemble_model = ensemble(models,model_input)

i also check similar issue.
i change that Model(inputs=[model_input,model_input,model_input],y,name='ensemble',it doesn't work.
Do you give some suggestions.
thanks.

It would be great if you can help letting know how to add a pretrained model in to this ensemble. I tried to add VGG16 through a model definition. However, vgg16 takes the input like this:
applications.vgg16(weights='imagenet', include_top=False, input_shape=(img_rows, img_cols, 3))
i'm not able to use model_input as you used in the definition file of all the custom models. it would be great if you can help how to add a pretrained model into this ensemble. You help is very much appreciated.
If you have the code, please pm me at [email protected]
Thanks once again.