| 1 | /**
|
| 2 | * @license
|
| 3 | * Copyright 2020 Google LLC. All Rights Reserved.
|
| 4 | * Licensed under the Apache License, Version 2.0 (the "License");
|
| 5 | * you may not use this file except in compliance with the License.
|
| 6 | * You may obtain a copy of the License at
|
| 7 | *
|
| 8 | * http://www.apache.org/licenses/LICENSE-2.0
|
| 9 | *
|
| 10 | * Unless required by applicable law or agreed to in writing, software
|
| 11 | * distributed under the License is distributed on an "AS IS" BASIS,
|
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 13 | * See the License for the specific language governing permissions and
|
| 14 | * limitations under the License.
|
| 15 | * =============================================================================
|
| 16 | */
|
| 17 | import { Tensor } from '../tensor';
|
| 18 | import { TensorLike } from '../types';
|
| 19 | /**
|
| 20 | * This operation reshapes the "batch" dimension 0 into `M + 1` dimensions of
|
| 21 | * shape `blockShape + [batch]`, interleaves these blocks back into the grid
|
| 22 | * defined by the spatial dimensions `[1, ..., M]`, to obtain a result with
|
| 23 | * the same rank as the input. The spatial dimensions of this intermediate
|
| 24 | * result are then optionally cropped according to `crops` to produce the
|
| 25 | * output. This is the reverse of `tf.spaceToBatchND`. See below for a precise
|
| 26 | * description.
|
| 27 | *
|
| 28 | * ```js
|
| 29 | * const x = tf.tensor4d([1, 2, 3, 4], [4, 1, 1, 1]);
|
| 30 | * const blockShape = [2, 2];
|
| 31 | * const crops = [[0, 0], [0, 0]];
|
| 32 | *
|
| 33 | * x.batchToSpaceND(blockShape, crops).print();
|
| 34 | * ```
|
| 35 | *
|
| 36 | * @param x A `tf.Tensor`. N-D with `x.shape` = `[batch] + spatialShape +
|
| 37 | * remainingShape`, where spatialShape has `M` dimensions.
|
| 38 | * @param blockShape A 1-D array. Must have shape `[M]`, all values must
|
| 39 | * be >= 1.
|
| 40 | * @param crops A 2-D array. Must have shape `[M, 2]`, all values must be >= 0.
|
| 41 | * `crops[i] = [cropStart, cropEnd]` specifies the amount to crop from input
|
| 42 | * dimension `i + 1`, which corresponds to spatial dimension `i`. It is required
|
| 43 | * that `cropStart[i] + cropEnd[i] <= blockShape[i] * inputShape[i + 1]`
|
| 44 | *
|
| 45 | * This operation is equivalent to the following steps:
|
| 46 | *
|
| 47 | * 1. Reshape `x` to `reshaped` of shape: `[blockShape[0], ...,
|
| 48 | * blockShape[M-1], batch / prod(blockShape), x.shape[1], ...,
|
| 49 | * x.shape[N-1]]`
|
| 50 | *
|
| 51 | * 2. Permute dimensions of `reshaped`to produce `permuted` of shape `[batch /
|
| 52 | * prod(blockShape),x.shape[1], blockShape[0], ..., x.shape[M],
|
| 53 | * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`
|
| 54 | *
|
| 55 | * 3. Reshape `permuted` to produce `reshapedPermuted` of shape `[batch /
|
| 56 | * prod(blockShape),x.shape[1] * blockShape[0], ..., x.shape[M] *
|
| 57 | * blockShape[M-1],x.shape[M+1], ..., x.shape[N-1]]`
|
| 58 | *
|
| 59 | * 4. Crop the start and end of dimensions `[1, ..., M]` of `reshapedPermuted`
|
| 60 | * according to `crops` to produce the output of shape: `[batch /
|
| 61 | * prod(blockShape),x.shape[1] * blockShape[0] - crops[0,0] - crops[0,1],
|
| 62 | * ..., x.shape[M] * blockShape[M-1] - crops[M-1,0] -
|
| 63 | * crops[M-1,1],x.shape[M+1], ..., x.shape[N-1]]`
|
| 64 | *
|
| 65 | * @doc {heading: 'Tensors', subheading: 'Transformations'}
|
| 66 | */
|
| 67 | declare function batchToSpaceND_<T extends Tensor>(x: T | TensorLike, blockShape: number[], crops: number[][]): T;
|
| 68 | export declare const batchToSpaceND: typeof batchToSpaceND_;
|
| 69 | export {};
|