1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
| /// <amd-module name="@tensorflow/tfjs-core/dist/ops/conv3d_backprop_input" />
| import { Tensor4D, Tensor5D } from '../tensor';
| /**
| * Computes the derivative of the input of a 3D convolution.
| *
| * @param xShape The shape of the input: [batch, depth, height, width,
| * in_channels]. If length of 4, batch of 1 is assumed.
| * @param dy The derivative of the output, of rank 5 or rank 4 of shape
| * `[batch, outDepth, outHeight, outWidth, in_channels]`.
| * If rank 4, batch of 1 is assumed.
| * @param filter The filter, rank 5, of shape
| * `[filterDepth, filterHeight, filterWidth, inDepth, outDepth]`.
| * @param strides The strides of the convolution: `[strideDepth, strideHeight,
| * strideWidth]`.
| * @param pad The type of padding algorithm used:
| * - `same` and stride 1: output will be of same size as input,
| * regardless of filter size.
| * - `valid`: output will be smaller than input if filter is larger
| * than 1x1.
| */
| declare function conv3DBackpropInput_<T extends Tensor4D | Tensor5D>(xShape: [
| number,
| number,
| number,
| number,
| number
| ] | [number, number, number, number], dy: T, filter: Tensor5D, strides: [number, number, number] | number, pad: 'valid' | 'same'): T;
| export declare const conv3DBackpropInput: typeof conv3DBackpropInput_;
| export {};
|
|
