ouwen / mimicknet Goto Github PK

would include author info, contact email, license, MimickNet name for package (?), make sure the version tracks your tag, etc.

Hello, how can I improve the resolution of single photos by using the pre-training model? Looking forward to your reply.Thanks.

https://github.com/KitwareMedical/SlicerITKUltrasound

It should be possible to add the preprocessing step into the graph. Users can also provide a flag to turn this feature off, but it would cut down on the boilerplate code required to run MimickNet.

maybe you mean the top one or left one out?

matlab_mimicknet_small_ladj.h5
mimicknet.h5
mimicknet_ladj.h5

Hi Sir,

In Python, below codes load the model "mimicknet_1568473738-210304.h5" successfully

mimicknet_model = tf.keras.models.load_model('./mimicknet_1568473738-210304.h5' compile=False)

However, if change the model file to 'padded_mimicknet_1568473738-210304.h5', it gives me the error :

ValueError: bad marshal data (unknown type code)

After some googling, that might be related to the Python version used when compiling the model file, though I believe you compiled these models under the same Python version (?), so for now I have no idea about how to proceed.

As my understanding, the "padded" version has built in the custom_pad() and custom_depad() function in layer input and output, so I can directly use the model without the codes to define custom_pad() and custom_depad(), so I would prefer to use the "padded" version, thanks and very appreciated for your help .

Hi Sir,

I believe below paragraph of Readme.me is incomplete, could you review this paragraph ? Thanks.

The model from the paper is provided as mimicknet_legacy.h5. We also provide a luminance adjusted version with fewer weights with mimicknet_1568473738-210304.h5 and padded_mimicknet_1568473738-210304.h5. The padded prefix allows for padding to be apart of the model, so no 16 padding logic is required. However, the layer used is incompatible with matlab, thus the matlab version has no pre-padding.

The mimicknet_phantom_verasonics.h5 models were generated with the

A notebook and sample data is provided under examples for use in the following environments:

matlab

unit tests needs to be written (issue #10 )

The first two calls to predict were slower than your quoted fps. You may want to put predict in a for loop to get to the timing on the 3rd or 4th frame...

Elapsed time is [ 6.941276, 0.356177, 0.090208, 0.094417 ] seconds.

Current citation points to the arxiv citation, inline and in the BibTeX at the end; that should be switched to the peer-reviewed TMI manuscript.