[GitHub] [incubator-mxnet] mazeltovlee opened a new issue #20951: MXNet fails to infer the shape of pad

[GitBox <gi...@apache.org>]

mazeltovlee opened a new issue #20951:
URL: https://github.com/apache/incubator-mxnet/issues/20951


   ## Description
   By constructing below symbolic graph:
   ```
   net = mx.symbol.Variable('data', shape=(0,10,10,1))
   net = mx.symbol.flatten(data=net)
   net = mx.sym.reshape(data=net, shape=(0, 10, 10, 1))
   net = mx.sym.pad(net, mode="constant", constant_value=0, pad_width=(0,0,0,0,1,1,1,1))
   ```
   and run `net.infer_shape_partial()`, mxnet will raise an error:
   ```
   [/usr/local/lib/python3.7/dist-packages/mxnet/base.py](https://localhost:8080/#) in check_call(ret)
       244     """
       245     if ret != 0:
   --> 246         raise get_last_ffi_error()
       247 
       248 
   
   MXNetError: MXNetError: Error in operator pad45: [15:10:16] ../src/operator/./pad-inl.h:215: Current implementation only supports 4-D or 5-D input.
   ```
   It seems to be an incorrect error since the input to pad is indeed a 4-D input.
   
   Moreover, if I delete the `flatten` operator, the `infer_shape_partial()` can work as expected:
   <img width="804" alt="image" src="https://user-images.githubusercontent.com/91937711/157691920-4c0f1dc9-4797-4f36-b9ec-35cbf56c8b3d.png">
   
   ## To Reproduce
   You can directly run above code to reproduce or access this colab link:
   https://colab.research.google.com/drive/1VXSTwD7O1DyGXR_Hp23lqcWapi7Xj1xN?usp=sharing
   


-- 
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.

To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org
For additional commands, e-mail: issues-help@mxnet.apache.org