maybeLee opened a new issue #20483: URL: https://github.com/apache/incubator-mxnet/issues/20483
## Description MXNet (version: 1.8.0) reports a crash when building a simple network containing three layers (Flatten, Reshape, and ZeroPadding2D), I use Keras as the front-end to build the model: ``` x = layers.Input(input_shape[1:]) layer_input = layers.Flatten()(x) layer_input = layers.Reshape((10,10,1))(layer_input) y = layers.ZeroPadding2D()(layer_input) model = Model(x, y) ``` However, if I use TensorFlow as the backend to run this model, I can get the result fluently. ### Error Message ``` Using MXNet backend /root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/__init__.py:31: DeprecationWarning: MXNet support in Keras is going to be discontinued and v2.2.4.3 is the last release as multi-backend Keras has been discontinued. It is recommended to consider switching to MXNet Gluon. More information can be found here: https://github.com/awslabs/keras-apache-mxnet "https://github.com/awslabs/keras-apache-mxnet";, DeprecationWarning) start building the model Traceback (most recent call last): File "try.py", line 20, in <module> y = layers.ZeroPadding2D()(layer_input) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/engine/base_layer.py", line 470, in __call__ output = self.call(inputs, **kwargs) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/layers/convolutional.py", line 2252, in call data_format=self.data_format) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 95, in func_wrapper train_symbol = func(*args, **kwargs) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 2361, in spatial_2d_padding x = _postprocess_convnd_output(x, data_format) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 87, in func_wrapper train_symbol = func(*args, **kwargs) File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 4803, in _postprocess_convnd_output if data_format == 'channels_last' and ndim(x) > 3: File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 536, in ndim shape = x.shape File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 4399, in shape return self._get_shape() File "/root/anaconda3/envs/diffcu_mxnet/lib/python3.6/site-packages/keras/backend/mxnet_backend.py", line 4408, in _get_shape _, out_shape, _ = self.symbol.infer_shape_partial() File "/mxnet/incubator-mxnet/python/mxnet/symbol/symbol.py", line 1177, in infer_shape_partial return self._infer_shape_impl(True, *args, **kwargs) File "/mxnet/incubator-mxnet/python/mxnet/symbol/symbol.py", line 1265, in _infer_shape_impl ctypes.byref(complete))) File "/mxnet/incubator-mxnet/python/mxnet/base.py", line 246, in check_call raise get_last_ffi_error() mxnet.base.MXNetError: MXNetError: Error in operator pad0: [08:36:54] src/operator/./pad-inl.h:216: Current implementation only supports 4-D or 5-D input. ``` ## To Reproduce You can reproduce this bug by running following code: ``` import os import argparse import sys import warnings parse = argparse.ArgumentParser() parse.add_argument("--bk", type=str,default="mxnet", help="the name of backend") flags, _ = parse.parse_known_args(sys.argv[1:]) os.environ["KERAS_BACKEND"]=flags.bk import keras from keras import initializers, layers, Model import numpy as np warnings.filterwarnings("ignore", category=DeprecationWarning) warnings.filterwarnings("ignore", category=UserWarning) input_shape = [1,10,10,1] print("start building the model") x = layers.Input(input_shape[1:]) layer_input = layers.Flatten()(x) layer_input = layers.Reshape((10,10,1))(layer_input) y = layers.ZeroPadding2D()(layer_input) print("successfully compile the model") model = Model(x, y) x = np.random.rand(**input_shape) pred = model.predict(x) print("successfully get the prediction result") ``` Run this script (suppose you store the code in the file: `try.py`) using backend MXNet, it will directly crash when compiling the model: ``` python try.py --bk mxnet ``` Run this script using backend TensorFlow, everything works fine: ``` python try.py --bk tensorflow ``` Related package version used: ``` mxnet==1.8.0 keras-mxnet==2.2.4.3 tensorflow==2.0.0 ``` It seems that this crash is caused by the one bug in the InferShape function when using ZeroPadding2D:  The rank of the input should be 4 dimensions but this function wrongly inferences the dimension. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
