Problem with conversion of YOLO model about coremltools HOT 6 CLOSED

Can you paste the repro script (instead of using the attachment) and some more detailed error logs here for us to debug more easily?

from coremltools.

Full output from console:

Conversion script:
coreml_model = coremltools.converters.keras.convert(
'yolo.h5',
input_names='image',
image_input_names='image',
output_names='grid',
image_scale=1/255.)

coreml_model.author = 'Redmon, Joseph and Farhadi, Ali'
coreml_model.license = 'Public Domain'
coreml_model.short_description = "Tiny YOLO neural network (2016), arXiv:1612.08242"

coreml_model.input_description['image'] = 'Input image'
coreml_model.output_description['grid'] = 'The 13x13 grid with the bounding box data'

print(coreml_model)
coreml_model.save('YOLO.mlmodel')

Structure of YOLO network

I'm using last coremtools available now (version 0.6.3). It's not possible to attach the Keras model here. So check my Google drive link in first post for the model.
Let me know if I can help with something else. Thank you.

from coremltools.

The console message indicates an error while loading the keras model. It looks like something is off while converting darknet configuration to keras model. Can you verify the validity of your Keras model?

from coremltools.

I'm new to the machine learning area so it's problematic for me to test it.
But the convertor from Darknet to Keras finished with success as with the TinyYOLO (simpler brother of YOLO). And I didn't have problem with converting TinyYOLO to mlmodel so that's why I thought the problem is in the coremltools.

from coremltools.

This looks like an issue with how the Keras model was saved. The error is happening when you are loading a Keras model from your .h5 file. I'm not sure this has to do with coremltools.

Please try the following:

coreml_model = coremltools.converters.keras.convert( keras_model, input_names='image', image_input_names='image', output_names='grid', image_scale=1/255.)

where keras_model is the model and not the filename

from coremltools.

@srikris You were right the problem is not in coremltools but something is wrong with the converted Keras model. I will try to investigate it further. I'm closing this issue.

Simple code proved problem in the model:

from keras.models import load_model
model = load_model('yolo-voc.h5')

from coremltools.