count_include_pad calculates the the most right output with dividing it by kernel size regardless it's not padded. #16203
Description
Describe the issue
From onnx/onnx#5276
count_include_pad calculates the most right output with dividing it by kernel size regardless it's not padded. I think it might be a bug that the original intention should be divided by the number of remaining pixels (including padding).
To reproduce
I am sorry, but it's easier to repro with onnx-script eager mode (backend ORT)
import numpy as np
import onnxscript
from onnxscript.onnx_opset import opset18 as op
x = np.array([[[ 2.0903, 4.6493, 1.6320, -3.2051, 4.6975, 4.7296, 3.3653,-1.5815, -2.3832],
[ 0.9628, -1.5899, -2.6820, 5.7529, 7.7346, -0.8910, -2.0151,0.1313, -0.5374]]]).astype(np.float32)
@onnxscript.script(default_opset=op)
def avg_pool(x):
result = op.AveragePool(x, kernel_shape=[7], strides=[3], pads=[3,3], ceil_mode=True, count_include_pad=True)
return result
print(avg_pool(x))
# [[[ 0.73807144 2.5655572 0.8032287 -0.08562858]
# [ 0.34911433 1.0389 1.4536142 -0.34588572]]]In PyTorch, it is divided by the number of remaining pixels (including padding):
import torch
a = torch.tensor([[[ 2.0903, 4.6493, 1.6320, -3.2051, 4.6975, 4.7296, 3.3653,-1.5815, -2.3832],[ 0.9628, -1.5899, -2.6820, 5.7529, 7.7346, -0.8910, -2.0151,0.1313, -0.5374]]])
p = torch.nn.AvgPool1d((7,), (3,), (3,), ceil_mode=True, count_include_pad=True)
p(a)
# tensor([[[ 0.7381, 2.5656, 0.8032, -0.0999],
# [ 0.3491, 1.0389, 1.4536, -0.4035]]])To summary, we can take a specific looke on the output of the first row. The reason that they are mis matched is because they are divided by different number of remaing pixels:
In the last window slide,
PyTorch: sum([3.3653,-1.5815, -2.3832, 0, 0, 0]) / 6 = -0.0999
ORT: sum([3.3653,-1.5815, -2.3832, 0, 0, 0]) / 7 = -0.08562858
Urgency
No response
Platform
Linux
OS Version
20.04.5
ONNX Runtime Installation
Built from Source
ONNX Runtime Version or Commit ID
NIghtly
ONNX Runtime API
Python
Architecture
X64
Execution Provider
Default CPU
Execution Provider Library Version
No response