KeyError: "The name 'embeddings:0' refers to a Tensor which does not exist. The operation, 'embeddings', does not exist in the graph."

[huangzicheng]

when i run validate_on_lfw.py, it always show that error.........i have used trian_softmax.py training results to run validate_on_lfw.py ， it was also....could give some help.....@davidsandberg

[raphasramos]

Did u solve this problem?

[StonePanda]

hello, did u solve this problem?

[petervickers]

Hi, this may be too late to be helpful but I found this error occured when I passed the index file of the saved graph instead of the directory itself. (see src/facenet.get_model_filenames)

[Vitomir2]

@davidsandberg Dear David,
I have the the error mentioned in the title. I am using the pre-defined model that is given in the 'Wiki' page and there is no 'embeddings' tensor in the default graph. I found what is the problem, the .meta file cannot imported to the graph and it gives the following error:

`` Node 'gradients/InceptionResnetV1/Bottleneck/BatchNorm/cond/FusedBatchNorm_1_grad/FusedBatchNormGrad' has an _output_shapes attribute inconsistent with the GraphDef for output #3: Dimension 0 in both shapes must be equal, but are 0 and 512. Shapes are [0] and [512]. ''

It gives the error in the facenet.py file, on line 381 - saver = tf.train.import_meta_graph(os.path.join(model_exp, meta_file), input_map=input_map)

Do you know how I can solve this issue?

I followed the 'Wiki' step by step and everything was okay, till the last step where I face this problem.

Best regards,
Vitomir