can't load the weights from weights_SSD300.hdf5

[xikaichen]

When I run SSD.py file. Fail to run this code : model.load_weights('weights_SSD300.hdf5', by_name=True) It gives error: ValueError: Dimension 0 in both shapes must be equal, but are 64 and 3 for 'Assign_4' (op: 'Assign') with input shapes: [64,300,3,3], [3,3,3,64].

Seems the weights_SSD300.hdf5 doesn't match your model. Could you please help me with this. Thank you very much.

[rykov8]

@xikaichen are you using the same setup as in README? Probsbly, something in Keras or Tensorflow has changed recently and that is the source of the problem.

[yukiB]

I also tried to run SSD.py file, however I failed to run the same code : model.load_weights('weights_SSD300.hdf5', by_name=True)

It gives error: ValueError: Dimension 0 in both shapes must be equal, but are 64 and 3

I used TensorFlow v0.11.0 and Keras v1.1.2 as README describes.

I seem the weights_SSD300.hdf5 doesn't match your model as @xikaichen mentioned in above. Could you please help me with this. Thank you in advance.

[xikaichen]

hi, try to use python3. It might work On Sat, Feb 18, 2017 at 2:26 AM Yuki BAN [email protected] wrote:

I also tried to run SSD.py file, however I failed to run the same code : model.load_weights('weights_SSD300.hdf5', by_name=True)

It gives error: ValueError: Dimension 0 in both shapes must be equal, but are 64 and 3

I used TensorFlow v0.11.0 and Keras v1.1.2 as README describes.

I seem the weights_SSD300.hdf5 doesn't match your model as @xikaichen https://github.com/xikaichen mentioned in above. Could you please help me with this. Thank you in advance.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/rykov8/ssd_keras/issues/37#issuecomment-280828382, or mute the thread https://github.com/notifications/unsubscribe-auth/AOJkoSuahmuI-InwzoQIKpzl9hE5tOGgks5rdp0HgaJpZM4L_bJH .

[yukiB]

It was my mistake of setting up of .keras/keras.json, thank you.

[anmolkalia]

What was the mistake you were making @yukiB? I am facing the same error.

[yukiB]

hi @anmolkalia,

Maybe your keras.json is like this

{
    "image_dim_ordering": "th",
    "epsilon": 1e-07,
    "floatx": "float32",
    "backend": "tensorflow"
}

If it's that, you should alter mage_dim_ordering to tf. Data's structure is different between th (channels, conv_dim1, conv_dim2, conv_dim3) and tf (conv_dim1, conv_dim2, conv_dim3, channels).

[FredlinT]

hi @anmolkalia @yukiB I meet a similar problem ,when i run a project using keras backened by tesorflow.When i load a weight .h5, it raised a error as follow:

ValueError: Dimension 0 in both shapes must be equal, but are 512 and 256 for 'Assign_12' (op: 'Assign') with input shapes: [512,768], [256,768].

I can't figure out .Could you please help me with is ,thanks!

[piotrbazan]

I have similar problem. I implemented various autoencoders from keras blog. Saving and restoring weights for NN model always works ok. However saving/loading for conv-net causes random errors: Can't load weights to model Dimension 0 in both shapes must be equal, but are 3 and 16 for 'Assign_2504' (op: 'Assign') with input shapes: [3,3,1,8], [16,1,3,3] What's interesting is that sometimes shapes are [3,3,1,8], [16,1,3,3] and sometimes [16,1,3,3][3,3,1,8]. Even more interesting is that rarerly it works without any error. Also I noticed that if I reuse existing model - loading weights works, still when creating model from scratch it causes errors. All code is run in Jupyter notebook within same jupyter session.

[yusuke0324]

I've got same error in loading weights, which are dumped from my own ssd codes for keras 2.1.5 and tf 1.4.

anyone got the solution?

Dimension 0 in both shapes must be equal, but are 3 and 63 for 'Assign_879' (op: 'Assign') with input shapes: [3,3,512,60], [63,512,3,3].

[baojunzhao]

@yusuke0324 I just got same error as yours. I just change the version to keras 2.1.4 & tensorflow-gpu 1.3.0. you can have a try