LpPool - 2 vs 11

Files changed (1) hide show
  1. LpPool2 → LpPool11 +8 -5
LpPool2 → LpPool11 RENAMED
@@ -1 +1 @@
1
1
  LpPool consumes an input tensor X and applies Lp pooling across
2
2
  the tensor according to kernel sizes, stride sizes, and pad lengths.
3
3
  Lp pooling consisting of computing the Lp norm on all values of a subset
4
4
  of the input tensor according to the kernel size and downsampling the
5
5
  data into the output tensor Y for further processing.
6
6
  **Attributes**
7
7
  * **auto_pad**:
8
8
  auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID.
9
9
  Where default value is NOTSET, which means explicit padding is used.
10
- SAME_UPPER or SAME_LOWER mean pad the input so that the output
10
+ SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i]
11
+ = ceil(input_shape[i] / strides[i]) for each axis i. The padding
12
+ is split between the two sides equally or almost equally (depending
11
- spatial size match the input.In case of odd number add the extra
13
+ on whether it is even or odd). In case the padding is an odd number,
12
- padding at the end for SAME_UPPER and at the beginning for
14
+ the extra padding is added at the end for SAME_UPPER and at the
13
- SAME_LOWER. VALID mean no padding.
15
+ beginning for SAME_LOWER.
14
16
  * **kernel_shape** (required):
15
17
  The size of the kernel along each axis.
16
18
  * **p**:
17
19
  p value of the Lp norm used to pool over the input data.
18
20
  * **pads**:
19
21
  Padding for the beginning and ending along each spatial axis, it can
20
22
  take any value greater than or equal to 0. The value represent the
21
23
  number of pixels added to the beginning and end part of the
22
24
  corresponding axis. pads format should be as follow [x1_begin,
23
25
  x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels
24
26
  added at the beginning of axis i and xi_end, the number of pixels
25
27
  added at the end of axis i. This attribute cannot be used
26
28
  simultaneously with auto_pad attribute. If not present, the padding
27
29
  defaults to 0 along start and end of each spatial axis.
28
30
  * **strides**:
31
+ Stride along each spatial axis. If not present, the stride defaults
29
- Stride along each spatial axis.
32
+ to 1 along each spatial axis.
30
33
  **Inputs**
31
34
  * **X** (heterogeneous) - **T**:
32
35
  Input data tensor from the previous operator; dimensions for image
33
36
  case are (N x C x H x W), where N is the batch size, C is the number
34
37
  of channels, and H and W are the height and the width of the data.
35
38
  For non image case, the dimensions are in the form of (N x C x D1 x
36
39
  D2 ... Dn), where N is the batch size.
37
40
  **Outputs**
38
41
  * **Y** (heterogeneous) - **T**:
39
42
  Output data tensor from Lp pooling across the input tensor.
40
43
  Dimensions will vary based on various kernel, stride, and pad sizes.
41
44
  **Type Constraints**
42
45
  * **T** in (
43
46
  tensor(double),
44
47
  tensor(float),
45
48
  tensor(float16)
46
49
  ):
47
50
  Constrain input and output types to float tensors.