[Global Pooling support] Convert Models from Caffe to Tensorflow about mmdnn

[kitstar]

Hi, Please refer the last step of Tensorflow ReadME and see if it works. Thanks.

Btw, we just tested caffe vgg and googlenet model (with inference result validation). More pre-trained model samples are welcome.

[seanchung2]

Hi,

My numpy files were converted by ethereon's converter. Since the output by his converter will include from kaffe.tensorflow import Network, they can't be fit into your converter.

So I tried to use your caffe to IR converter. The error happened as following:

Traceback (most recent call last):
  File "/usr/lib/python2.7/runpy.py", line 174, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 156, in <module>
    _main()
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 151, in _main
    ret = _convert(args)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 10, in _convert
    graph = transformer.transform_graph()
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/transformer.py", line 348, in transform_graph
    mapped_node = self.map_node(node)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/transformer.py", line 368, in map_node
    mapped_node = map_func(node)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/mapper.py", line 117, in map_pooling
    kwargs, padding = cls.get_kernel_params(node)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/mapper.py", line 38, in get_kernel_params
    if node.kernel_parameters.p_h > 0 or node.kernel_parameters.p_w > 0:
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/graph.py", line 187, in kernel_parameters
    k_h = self.get_kernel_value(params.kernel_h, params.kernel_size, 0)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/graph.py", line 180, in get_kernel_value
    raise ValueError('Unable to determine kernel parameter!')
ValueError: Unable to determine kernel parameter!

Can you please check it out?
If you wanna reproduce the error, here is my caffe model: Xception

Thank you.

[kitstar]

@seanchung2 Checked the model. Currently we don't support Conv with groups attribute. We will implement it soon. Thanks.

[kitstar]

Hi @seanchung2 ,
The problem is lacking support of Global Pooling in old version MMdnn. Newest master implement it and tested in my env. Please have a try.
Thanks.