mindspore.nn.MaxPool2d throws exception but shows MaxPool3D

[PhyllisJi]

Environment

Hardware Environment(Ascend/GPU/CPU): GPU

Software Environment:

• MindSpore version (source or binary): 2.2.14
• Python version (e.g., Python 3.7.5): 3.8
• OS platform and distribution (e.g., Linux Ubuntu 16.04):
• GCC/Compiler version (if compiled from source):

Describe the current behavior

It should be mindspore.nn.MaxPool2d throwing exceptions, but the display is MaxPool3D, which looks misleading.

Steps to reproduce the issue

import mindspore
import numpy as np


class Model_ms(mindspore.nn.Cell):
    def __init__(self):
        super(Model_ms, self).__init__()
        self.conv1_mutated = mindspore.nn.Conv2d(in_channels=1, out_channels=6, kernel_size=(8, 8), stride=(1, 1), pad_mode="pad", padding=(0, 0, 0, 0), dilation=(1, 1), group=1, has_bias=True)
        self.relu1_mutated = mindspore.nn.HSwish()
        self.pool1 = mindspore.nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2), pad_mode="pad", padding=(0, 0), dilation=1, return_indices=False, ceil_mode=False)

    def construct(self, input):
        conv1_output = self.conv1_mutated(input)
        relu1_output = self.relu1_mutated(conv1_output)
        maxpool1_output = self.pool1(relu1_output)

        tail_fc_output = maxpool1_output
        return tail_fc_output


inp = np.random.random([1, 1, 8, 23]).astype(np.float32)
ms_model = Model_ms()
ms_input = mindspore.Tensor(inp)
ms_output = ms_model(ms_input)
print(ms_output)

Related log / screenshot

ValueError: for MaxPool3DWithArgmax, shape of out is [1, 6, 1, 0, 8]. It should be not less than zero.

---

• C++ Call Stack: (For framework developers)

---

mindspore/core/ops/max_pool3d_with_argmax.cc:266 MaxPool3DWithArgmaxInferShape

[zhouyifeng888]

The reason for the error here is that the shape is [1, 6, 1, 0, 8], where the 0 in the fourth dimension is an invalid input; it must be greater than 0 to be correct. The error message from MaxPool3d is reported because the underlying implementation of mindspore.nn.MaxPool2d uses the maxpool3dwithargmax operator.

[PhyllisJi]

The reason for the error here is that the shape is [1, 6, 1, 0, 8], where the 0 in the fourth dimension is an invalid input; it must be greater than 0 to be correct. The error message from MaxPool3d is reported because the underlying implementation of mindspore.nn.MaxPool2d uses the maxpool3dwithargmax operator.

Thank you for your reply! I think that's the case, mindspore should be catching exceptions at a higher level of code so that it doesn't throw confusing exception hints.