wangliting-amy opened a new issue #6564:
URL: https://github.com/apache/incubator-tvm/issues/6564
hello:
i want to build and run the model of resnest50,which i get_model from mxnet
model_zoo.the code is:
model = gluoncv.model_zoo.get_model('ResNeSt50', pretrained=True)
and i want to change the layout from NCHW to NHWC, but i find the
adaptive_avg_pool2d can not be changed the layout.
so according to the link of
https://tvm.apache.org/docs/dev/convert_layout.html?highlight=convertlayout
i add the FTVMConvertLayout register for adaptive_avg_pool2d, the code like
this:
@reg.register_convert_op_layout("nn.adaptive_avg_pool2d")
def convert_adaptive_avg_pool2d(attrs, inputs, tinfos, desired_layouts):
"""Convert Layout pass registration for adaptive_avg_pool2d op.
Parameters
----------
attrs : tvm.ir.Attrs
Attributes of current convolution
inputs : list of tvm.relay.Expr
The args of the Relay expr to be legalized
tinfos : list of types
List of input and output types
desired_layouts : list of layout strings
List of layouts defining our desired
layout for the data input.
Returns
-------
result : tvm.relay.Expr
The transformed expr
"""
# pylint: disable=import-outside-toplevel
from tvm import relay
data = inputs
new_attrs = dict(attrs)
assert len(desired_layouts) == 1, "A desired layout is expected for
input data only"
desired_data_layout = desired_layouts[0]
assert desired_data_layout != "default", "Data layout cannot be default"
new_attrs['layout'] = desired_data_layout
return relay.nn.adaptive_avg_pool2d(data, **new_attrs)
but when i build and run,there is a problem like this:
/tvm/python/tvm/relay/op/nn/_nn.py", line 420, in convert_adaptive_avg_pool2d
new_attrs = dict(attrs)
TypeError: 'Object' object is not iterable
i do not know the reason,can help me,thank you~
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
