Conv - 1 vs 11

Next section compares an older to a newer version of the same operator after both definition are converted into markdown text.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. Conv1 → Conv11 +9 -16

Conv1 → Conv11 RENAMED Viewed

	@@ -1 +1 @@
1 1	The convolution operator consumes an input tensor and a filter, and
2 2	computes the output.
3 3	**Attributes**
4 4	* **auto_pad**:
5 5	auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID.
6 6	Where default value is NOTSET, which means explicit padding is used.
7	- SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i]
7	+ SAME_UPPER or SAME_LOWER mean pad the input so that the output
8	+ spatial size match the input.In case of odd number add the extra
9	+ padding at the end for SAME_UPPER and at the beginning for
10	+ SAME_LOWER. VALID mean no padding.
8	- = ceil(input_shape[i] / strides[i]) for each axis i. The padding
9	- is split between the two sides equally or almost equally (depending
10	- on whether it is even or odd). In case the padding is an odd number,
11	- the extra padding is added at the end for SAME_UPPER and at the
12	- beginning for SAME_LOWER.
13 11	* **dilations**:
14	- dilation value along each spatial axis of the filter. If not
12	+ dilation value along each spatial axis of the filter.
15	- present, the dilation defaults is 1 along each spatial axis.
16 13	* **group**:
17 14	number of groups input channels and output channels are divided
18 15	into.
19 16	* **kernel_shape**:
20 17	The shape of the convolution kernel. If not present, should be
21 18	inferred from input W.
22 19	* **pads**:
23 20	Padding for the beginning and ending along each spatial axis, it can
24 21	take any value greater than or equal to 0. The value represent the
25 22	number of pixels added to the beginning and end part of the
26 23	corresponding axis. pads format should be as follow [x1_begin,
27 24	x2_begin...x1_end, x2_end,...], where xi_begin the number of pixels
28 25	added at the beginning of axis i and xi_end, the number of pixels
29 26	added at the end of axis i. This attribute cannot be used
30 27	simultaneously with auto_pad attribute. If not present, the padding
31 28	defaults to 0 along start and end of each spatial axis.
32 29	* **strides**:
33	- Stride along each spatial axis. If not present, the stride defaults
34	- is 1 along each spatial axis.
30	+ Stride along each spatial axis.
35 31	**Inputs**
36 32	Between 2 and 3 inputs.
37 33	* **X** (heterogeneous) - **T**:
38 34	Input data tensor from previous layer; has size (N x C x H x W),
39 35	where N is the batch size, C is the number of channels, and H and W
40 36	are the height and width. Note that this is for the 2D image.
41 37	Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if
42 38	dimension denotation is in effect, the operation expects input data
43 39	tensor to arrive with the dimension denotation of [DATA_BATCH,
44 40	DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].
45 41	* **W** (heterogeneous) - **T**:
46 42	The weight tensor that will be used in the convolutions; has size (M
47 43	x C/group x kH x kW), where C is the number of channels, and kH and
48 44	kW are the height and width of the kernel, and M is the number of
49 45	feature maps. For more than 2 dimensions, the kernel shape will be
50 46	(M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the
51 47	dimension of the kernel. Optionally, if dimension denotation is in
52 48	effect, the operation expects the weight tensor to arrive with the
53 49	dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL,
50	+ FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] *
51	+ group) == C (assuming zero based indices for the shape array). Or in
52	+ other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
54	- FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for
55	- the shape array, X.shape[1] == (W.shape[1] * group) == C and
56	- W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL
57	- multiplied by the number of groups should be equal to DATA_CHANNEL
58	- and the number of feature maps M should be a multiple of the number
59	- of groups G.
60 53	* **B** (optional, heterogeneous) - **T**:
61 54	Optional 1D bias to be added to the convolution, has size of M.
62 55	**Outputs**
63 56	* **Y** (heterogeneous) - **T**:
64 57	Output data tensor that contains the result of the convolution. The
65 58	output dimensions are functions of the kernel size, stride size, and
66 59	pad lengths.
67 60	**Type Constraints**
68 61	* **T** in (
69 62	tensor(double),
70 63	tensor(float),
71 64	tensor(float16)
72 65	):
73 66	Constrain input and output types to float tensors.