BatchNormalization - 6 vs 7#

Next section compares an older to a newer version of the same operator after both definition are converted into markdown text. Green means an addition to the newer version, red means a deletion. Anything else is unchanged.

Files changed (1) hide show
  1. BatchNormalization6 → BatchNormalization7 +25 -22
BatchNormalization6 → BatchNormalization7 RENAMED
@@ -1 +1 @@
1
1
  Carries out batch normalization as described in the paper
2
2
  https://arxiv.org/abs/1502.03167. Depending on the mode it is being run,
3
3
  there are multiple cases for the number of outputs, which we list below:
4
4
  Output case #1: Y, mean, var, saved_mean, saved_var (training mode)
5
5
  Output case #2: Y (test mode)
6
- This operator has **optional** inputs/outputs. See ONNX <https://github.com/onnx/onnx/blob/master/docs/IR.md>_ for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument's name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
7
6
  **Attributes**
8
7
  * **epsilon**:
9
- The epsilon value to use to avoid division by zero.
8
+ The epsilon value to use to avoid division by zero, default is
9
+ 1e-5f.
10
+ * **is_test**:
11
+ If set to nonzero, run spatial batch normalization in test mode,
12
+ default is 0.
10
13
  * **momentum**:
11
14
  Factor used in computing the running mean and variance.e.g.,
12
- running_mean = running_mean * momentum + mean * (1 - momentum).
15
+ running_mean = running_mean * momentum + mean * (1 - momentum),
16
+ default is 0.9f.
13
17
  * **spatial**:
14
- If true, compute the mean and variance across per activation. If
18
+ If true, compute the mean and variance across all spatial elements
15
- false, compute the mean and variance across per feature over each
19
+ If false, compute the mean and variance across per feature.Default
16
- mini-batch.
20
+ is 1.
17
21
  **Inputs**
18
22
  * **X** (heterogeneous) - **T**:
19
23
  Input data tensor from the previous operator; dimensions for image
20
24
  case are (N x C x H x W), where N is the batch size, C is the number
21
25
  of channels, and H and W are the height and the width of the data.
22
26
  For non image case, the dimensions are in the form of (N x C x D1 x
23
27
  D2 ... Dn), where N is the batch size.
24
28
  * **scale** (heterogeneous) - **T**:
25
- If spatial is true, the dimension of scale is (C). If spatial is
26
- false, the dimensions of scale are (C x D1 x ... x Dn)
29
+ The scale as a 1-dimensional tensor of size C to be applied to the
30
+ output.
27
31
  * **B** (heterogeneous) - **T**:
28
- If spatial is true, the dimension of bias is (C). If spatial is
29
- false, the dimensions of bias are (C x D1 x ... x Dn)
32
+ The bias as a 1-dimensional tensor of size C to be applied to the
33
+ output.
30
34
  * **mean** (heterogeneous) - **T**:
31
- If spatial is true, the dimension of the running mean (training) or
32
- the estimated mean (testing) is (C). If spatial is false, the
33
- dimensions of the running mean (training) or the estimated mean
35
+ The running mean (training) or the estimated mean (testing) as a
34
- (testing) are (C x D1 x ... x Dn).
36
+ 1-dimensional tensor of size C.
35
37
  * **var** (heterogeneous) - **T**:
38
+ The running variance (training) or the estimated variance (testing)
39
+ as a 1-dimensional tensor of size C.
36
- If spatial is true, the dimension of the running variance(training)
37
- or the estimated variance (testing) is (C). If spatial is false, the
38
- dimensions of the running variance(training) or the estimated
39
- variance (testing) are (C x D1 x ... x Dn).
40
40
  **Outputs**
41
41
  Between 1 and 5 outputs.
42
42
  * **Y** (heterogeneous) - **T**:
43
- The output tensor of the same shape as X
43
+ The output tensor of the same shape as X.
44
44
  * **mean** (optional, heterogeneous) - **T**:
45
- The running mean after the BatchNormalization operator.
45
+ The running mean after the BatchNormalization operator. Must be in-
46
+ place with the input mean. Should not be used for testing.
46
47
  * **var** (optional, heterogeneous) - **T**:
47
- The running variance after the BatchNormalization operator.
48
+ The running variance after the BatchNormalization operator. Must be
49
+ in-place with the input var. Should not be used for testing.
48
50
  * **saved_mean** (optional, heterogeneous) - **T**:
49
51
  Saved mean used during training to speed up gradient computation.
52
+ Should not be used for testing.
50
53
  * **saved_var** (optional, heterogeneous) - **T**:
51
54
  Saved variance used during training to speed up gradient
52
- computation.
55
+ computation. Should not be used for testing.
53
56
  **Type Constraints**
54
57
  * **T** in (
55
58
  tensor(double),
56
59
  tensor(float),
57
60
  tensor(float16)
58
61
  ):
59
62
  Constrain input and output types to float tensors.