Exporting the pre-training model failed

[surmo]

I am use this code to export the pre-model,but it fail

net = gcv.model_zoo.get_model('psp_resnet101_ade', pretrained=True)
export_block('psp_resnet101_ade', net,data_shape=(224, 224, 3), preprocess=True, layout='HWC')

Traceback (most recent call last):
File "I:/Python_project/MachineLearninginAction/exportModel.py", line 6, in
export_block('psp_resnet101_ade', net,data_shape=(224, 224, 3), preprocess=None, layout='HWC')
File "C:\Users\Sumo\AppData\Roaming\Python\Python35\site-packages\gluoncv\utils\export_helper.py", line 109, in export_block
wrapper_block(x)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 542, in call
out = self.forward(*args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 909, in forward
return self._call_cached_op(x, *args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 799, in _call_cached_op
self._build_cache(*args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 751, in _build_cache
data, out = self._get_graph(*args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 743, in _get_graph
out = self.hybrid_forward(symbol, *grouped_inputs, **params) # pylint: disable=no-value-for-parameter
File "C:\Users\Sumo\AppData\Roaming\Python\Python35\site-packages\gluoncv\model_zoo\pspnet.py", line 53, in hybrid_forward
x = self.head(c4)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 542, in call
out = self.forward(*args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 926, in forward
return self.hybrid_forward(symbol, x, *args, **params)
File "C:\Users\Sumo\AppData\Roaming\Python\Python35\site-packages\gluoncv\model_zoo\pspnet.py", line 115, in hybrid_forward
x = self.psp(x)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 542, in call
out = self.forward(*args)
File "C:\Users\Sumo\AppData\Local\Programs\Python\Python35\lib\site-packages\mxnet\gluon\block.py", line 926, in forward
return self.hybrid_forward(symbol, x, *args, **params)
File "C:\Users\Sumo\AppData\Roaming\Python\Python35\site-packages\gluoncv\model_zoo\pspnet.py", line 91, in hybrid_forward
_, _, h, w = x.shape
AttributeError: 'Symbol' object has no attribute 'shape'

[zhreshold]

I identified several segmentation models are not hybridizable now, since the models use shape inside the graph. For now only FCN models can be exported without issue.

The workaround for PSP and DeepLab models are to remove the line:

gluon-cv/gluoncv/model_zoo/pspnet.py

Line 91 in 6517ee4

_, _, h, w = x.shape

and use fixed shape (e.g., h=w=448)

[zhreshold]

We are adding it as feature request duplicate in #547 , I am closing this for now, feel free to join the discussion in #547