com.simiacryptus.mindseye.layers.java.ImgReshapeLayer Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of mindseye-java Show documentation
Show all versions of mindseye-java Show documentation
Pure Java Neural Networks Components
/*
* Copyright (c) 2019 by Andrew Charneski.
*
* The author licenses this file to you under the
* Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance
* with the License. You may obtain a copy
* of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package com.simiacryptus.mindseye.layers.java;
import com.google.gson.JsonObject;
import com.simiacryptus.mindseye.lang.*;
import javax.annotation.Nonnull;
import javax.annotation.Nullable;
import java.util.*;
import java.util.stream.IntStream;
/**
* Reduces or expands png resolution by rearranging the values in NxN tiles to effectively stripe the small-scale
* spacial dimension across N^2 color bands.
*/
@SuppressWarnings("serial")
public class ImgReshapeLayer extends LayerBase {
private final boolean expand;
private final int kernelSizeX;
private final int kernelSizeY;
/**
* Instantiates a new Img reshapeCast key.
*
* @param kernelSizeX the kernel size x
* @param kernelSizeY the kernel size y
* @param expand the expandPlasma
*/
public ImgReshapeLayer(final int kernelSizeX, final int kernelSizeY, final boolean expand) {
super();
this.kernelSizeX = kernelSizeX;
this.kernelSizeY = kernelSizeY;
this.expand = expand;
}
/**
* Instantiates a new Img reshapeCast key.
*
* @param json the json
*/
protected ImgReshapeLayer(@Nonnull final JsonObject json) {
super(json);
kernelSizeX = json.getAsJsonPrimitive("kernelSizeX").getAsInt();
kernelSizeY = json.getAsJsonPrimitive("kernelSizeY").getAsInt();
expand = json.getAsJsonPrimitive("expandPlasma").getAsBoolean();
}
/**
* Copy condense tensor.
*
* @param inputData the input data
* @param outputData the output data
* @return the tensor
*/
@Nonnull
public static Tensor copyCondense(@Nonnull final Tensor inputData, @Nonnull final Tensor outputData) {
@Nonnull final int[] inDim = inputData.getDimensions();
@Nonnull final int[] outDim = outputData.getDimensions();
assert 3 == inDim.length;
assert 3 == outDim.length;
assert inDim[0] >= outDim[0];
assert inDim[1] >= outDim[1];
assert inDim[2] < outDim[2];
assert 0 == inDim[0] % outDim[0];
assert 0 == inDim[1] % outDim[1];
final int kernelSizeX = inDim[0] / outDim[0];
final int kernelSizeY = inDim[0] / outDim[0];
int index = 0;
@Nullable final double[] outputDataData = outputData.getData();
for (int z = 0; z < inDim[2]; z++) {
for (int xx = 0; xx < kernelSizeX; xx++) {
for (int yy = 0; yy < kernelSizeY; yy++) {
for (int y = 0; y < inDim[1]; y += kernelSizeY) {
for (int x = 0; x < inDim[0]; x += kernelSizeX) {
outputDataData[index++] = inputData.get(x + xx, y + yy, z);
}
}
}
}
}
return outputData;
}
/**
* Copy expandPlasma tensor.
*
* @param inputData the input data
* @param outputData the output data
* @return the tensor
*/
@Nonnull
public static Tensor copyExpand(@Nonnull final Tensor inputData, @Nonnull final Tensor outputData) {
@Nonnull final int[] inDim = inputData.getDimensions();
@Nonnull final int[] outDim = outputData.getDimensions();
assert 3 == inDim.length;
assert 3 == outDim.length;
assert inDim[0] <= outDim[0];
assert inDim[1] <= outDim[1];
assert inDim[2] > outDim[2];
assert 0 == outDim[0] % inDim[0];
assert 0 == outDim[1] % inDim[1];
final int kernelSizeX = outDim[0] / inDim[0];
final int kernelSizeY = outDim[0] / inDim[0];
int index = 0;
for (int z = 0; z < outDim[2]; z++) {
for (int xx = 0; xx < kernelSizeX; xx++) {
for (int yy = 0; yy < kernelSizeY; yy++) {
for (int y = 0; y < outDim[1]; y += kernelSizeY) {
for (int x = 0; x < outDim[0]; x += kernelSizeX) {
outputData.set(x + xx, y + yy, z, inputData.getData()[index++]);
}
}
}
}
}
return outputData;
}
/**
* From json img reshapeCast key.
*
* @param json the json
* @param rs the rs
* @return the img reshapeCast key
*/
public static ImgReshapeLayer fromJson(@Nonnull final JsonObject json, Map rs) {
return new ImgReshapeLayer(json);
}
@Nonnull
@Override
public Result eval(@Nonnull final Result... inObj) {
//assert Arrays.stream(inObj).flatMapToDouble(input-> input.getData().stream().flatMapToDouble(x-> Arrays.stream(x.getData()))).allMatch(v->Double.isFinite(v));
Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
final Result input = inObj[0];
final TensorList batch = input.getData();
@Nonnull final int[] inputDims = batch.getDimensions();
assert 3 == inputDims.length;
assert expand || 0 == inputDims[0] % kernelSizeX : (inputDims[0] + " % " + kernelSizeX);
assert expand || 0 == inputDims[1] % kernelSizeY : (inputDims[1] + " % " + kernelSizeY);
assert !expand || 0 == inputDims[2] % (kernelSizeX * kernelSizeY);
//assert input.getData().stream().flatMapToDouble(x-> Arrays.stream(x.getData())).allMatch(v->Double.isFinite(v));
Tensor outputDims;
if (expand) {
outputDims = new Tensor(inputDims[0] * kernelSizeX,
inputDims[1] * kernelSizeY,
inputDims[2] / (kernelSizeX * kernelSizeY));
} else {
outputDims = new Tensor(inputDims[0] / kernelSizeX,
inputDims[1] / kernelSizeY,
inputDims[2] * kernelSizeX * kernelSizeY);
}
TensorArray data = TensorArray.wrap(IntStream.range(0, batch.length()).parallel()
.mapToObj(dataIndex -> {
Tensor inputData = batch.get(dataIndex);
Tensor tensor = expand ? ImgReshapeLayer.copyExpand(inputData, outputDims.copy()) : ImgReshapeLayer.copyCondense(inputData, outputDims.copy());
inputData.freeRef();
return tensor;
})
.toArray(i -> new Tensor[i]));
outputDims.freeRef();
return new Result(data, (@Nonnull final DeltaSet buffer, @Nonnull final TensorList error) -> {
//assert error.stream().flatMapToDouble(x-> Arrays.stream(x.getData())).allMatch(v->Double.isFinite(v));
if (input.isAlive()) {
@Nonnull TensorArray tensorArray = TensorArray.wrap(IntStream.range(0, error.length()).parallel()
.mapToObj(dataIndex -> {
@Nonnull final Tensor passback = new Tensor(inputDims);
@Nullable final Tensor err = error.get(dataIndex);
Tensor tensor = expand ? ImgReshapeLayer.copyCondense(err, passback) : ImgReshapeLayer.copyExpand(err, passback);
err.freeRef();
return tensor;
}).toArray(i -> new Tensor[i]));
input.accumulate(buffer, tensorArray);
}
}) {
@Override
protected void _free() {
Arrays.stream(inObj).forEach(nnResult -> nnResult.freeRef());
}
@Override
public boolean isAlive() {
return input.isAlive() || !isFrozen();
}
};
}
@Nonnull
@Override
public JsonObject getJson(Map resources, DataSerializer dataSerializer) {
@Nonnull final JsonObject json = super.getJsonStub();
json.addProperty("kernelSizeX", kernelSizeX);
json.addProperty("kernelSizeY", kernelSizeX);
json.addProperty("expandPlasma", expand);
return json;
}
@Nonnull
@Override
public List state() {
return new ArrayList<>();
}
}