Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance.
Project price only 1 $
You can buy this project and download/modify it how often you want.
org.deeplearning4j.nn.layers.convolution.Convolution3DLayer Maven / Gradle / Ivy
/*
* ******************************************************************************
* *
* *
* * This program and the accompanying materials are made available under the
* * terms of the Apache License, Version 2.0 which is available at
* * https://www.apache.org/licenses/LICENSE-2.0.
* *
* * See the NOTICE file distributed with this work for additional
* * information regarding copyright ownership.
* * 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.
* *
* * SPDX-License-Identifier: Apache-2.0
* *****************************************************************************
*/
package org.deeplearning4j.nn.layers.convolution;
import org.deeplearning4j.exception.DL4JInvalidInputException;
import org.deeplearning4j.nn.conf.ConvolutionMode;
import org.deeplearning4j.nn.conf.NeuralNetConfiguration;
import org.deeplearning4j.nn.conf.layers.Convolution3D;
import org.deeplearning4j.nn.gradient.DefaultGradient;
import org.deeplearning4j.nn.gradient.Gradient;
import org.deeplearning4j.nn.params.Convolution3DParamInitializer;
import org.deeplearning4j.nn.workspace.ArrayType;
import org.deeplearning4j.nn.workspace.LayerWorkspaceMgr;
import org.deeplearning4j.util.Convolution3DUtils;
import org.nd4j.linalg.activations.IActivation;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.common.primitives.Pair;
import java.util.Arrays;
public class Convolution3DLayer extends ConvolutionLayer {
public Convolution3DLayer(NeuralNetConfiguration conf, DataType dataType) {
super(conf, dataType);
}
@Override
void initializeHelper() {
// no op
}
@Override
public Pair backpropGradient(INDArray epsilon, LayerWorkspaceMgr workspaceMgr) {
if (input.rank() != 5) {
throw new DL4JInvalidInputException("Got rank " + input.rank()
+ " array as input to SubsamplingLayer with shape " + Arrays.toString(input.shape())
+ ". Expected rank 5 array with shape [minibatchSize, channels, "
+ "inputHeight, inputWidth, inputDepth]. "
+ layerId());
}
INDArray input = this.input.castTo(dataType);
INDArray weights = getParamWithNoise(Convolution3DParamInitializer.WEIGHT_KEY, true, workspaceMgr);
Convolution3D layerConfig = (Convolution3D) layerConf();
boolean isNCDHW = layerConfig.getDataFormat() == Convolution3D.DataFormat.NCDHW;
long miniBatch = input.size(0);
int inD = (int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));
int inW = (int) (isNCDHW ? input.size(4) : input.size(3));
int outEpsChannels = (int) layerConf().getNIn();
int[] dilation = layerConfig.getDilation();
int[] kernel = layerConfig.getKernelSize();
int[] strides = layerConfig.getStride();
int[] pad;
int[] outSize;
if (convolutionMode == ConvolutionMode.Same) {
outSize = Convolution3DUtils.get3DOutputSize(
input, kernel, strides, null, convolutionMode, dilation, isNCDHW);
pad = Convolution3DUtils.get3DSameModeTopLeftPadding(
outSize, new int[]{inD, inH, inW}, kernel, strides, dilation);
} else {
pad = layerConfig.getPadding();
}
INDArray weightGradView = gradientViews.get(Convolution3DParamInitializer.WEIGHT_KEY);
INDArray outEpsilon = workspaceMgr.createUninitialized(ArrayType.ACTIVATION_GRAD, weights.dataType(),
miniBatch * outEpsChannels * inD * inH * inW);
if (isNCDHW)
outEpsilon = outEpsilon.reshape('c', miniBatch, outEpsChannels, inD, inH, inW);
else
outEpsilon = outEpsilon.reshape('c', miniBatch, inD, inH, inW, outEpsChannels);
int[] intArgs = new int[]{
kernel[0], kernel[1], kernel[2],
strides[0], strides[1], strides[2],
pad[0], pad[1], pad[2],
dilation[0], dilation[1], dilation[2],
convolutionMode == ConvolutionMode.Same ? 1 : 0,
isNCDHW ? 0 : 1
};
INDArray delta;
IActivation activation = layerConfig.getActivationFn();
Pair p = preOutput(true, true, workspaceMgr);
delta = activation.backprop(p.getFirst(), epsilon).getFirst();
INDArray bias;
INDArray biasGradView = null;
//DL4J conv3d weights: val weightsShape = new long[]{outputDepth, inputDepth, kernel[0], kernel[1], kernel[2]};
//libnd4j conv3d weights: [kD, kH, kW, iC, oC]
weights = weights.permute(2, 3, 4, 1, 0);
INDArray opWeightGradView = weightGradView.permute(2, 3, 4, 1, 0);
INDArray[] inputs;
INDArray[] outputs;
if (layerConfig.hasBias()) {
biasGradView = gradientViews.get(Convolution3DParamInitializer.BIAS_KEY);
bias = getParamWithNoise(Convolution3DParamInitializer.BIAS_KEY, true, workspaceMgr);
inputs = new INDArray[]{input, weights, bias, delta};
outputs = new INDArray[]{outEpsilon, opWeightGradView, biasGradView};
} else {
inputs = new INDArray[]{input, weights, delta};
outputs = new INDArray[]{outEpsilon, opWeightGradView};
}
CustomOp op = DynamicCustomOp.builder("conv3dnew_bp")
.addInputs(inputs)
.addIntegerArguments(intArgs)
.addOutputs(outputs)
.callInplace(false)
.build();
Nd4j.getExecutioner().exec(op);
Gradient retGradient = new DefaultGradient();
if (layerConfig.hasBias()) {
retGradient.setGradientFor(Convolution3DParamInitializer.BIAS_KEY, biasGradView);
}
retGradient.setGradientFor(Convolution3DParamInitializer.WEIGHT_KEY, weightGradView, 'c');
weightNoiseParams.clear();
return new Pair<>(retGradient, outEpsilon);
}
@Override
public INDArray preOutput(boolean training, LayerWorkspaceMgr workspaceMgr) {
return preOutput(training, false, workspaceMgr).getFirst();
}
protected Pair preOutput(boolean training, boolean forBackprop, LayerWorkspaceMgr workspaceMgr) {
Convolution3D layerConfig = (Convolution3D) layerConf();
ConvolutionMode mode = layerConfig.getConvolutionMode();
boolean isNCDHW = layerConfig.getDataFormat() == Convolution3D.DataFormat.NCDHW;
INDArray input = this.input.castTo(dataType);
INDArray weights = getParamWithNoise(Convolution3DParamInitializer.WEIGHT_KEY, training, workspaceMgr);
if (input.rank() != 5) {
String layerName = conf.getLayer().getLayerName();
if (layerName == null)
layerName = "(not named)";
throw new DL4JInvalidInputException("Got rank " + input.rank()
+ " array as input to Convolution3DLayer (layer name = " + layerName + ", layer index = "
+ index + ") with shape " + Arrays.toString(input.shape()) + ". "
+ "Expected rank 5 array with shape [minibatchSize, numChannels, inputHeight, "
+ "inputWidth, inputDepth]."
+ (input.rank() == 2
? " (Wrong input type (see InputType.convolutionalFlat()) or wrong data type?)"
: "")
+ " " + layerId());
}
long miniBatch = input.size(0);
int inputChannels = (int) (isNCDHW ? input.size(1) : input.size(4));
int inD =(int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));
int inW = (int) (isNCDHW ? input.size(4) : input.size(3));
int outWeightChannels = (int)layerConf().getNOut();
int inWeightChannels = (int)layerConf().getNIn();
if (inputChannels != inWeightChannels) {
String layerName = conf.getLayer().getLayerName();
if (layerName == null)
layerName = "(not named)";
long dataInCh = isNCDHW ? input.size(1) : input.size(4);
String df;
if(isNCDHW){
df = ", dataFormat=NCDHW, [minibatch, inputChannels, depth, height, width]=";
} else {
df = ", dataFormat=NDHWC, [minibatch, depth, height, width, inputChannels]=";
}
throw new DL4JInvalidInputException("Cannot do forward pass in Convolution3D layer (layer name = "
+ layerName
+ ", layer index = " + index + "): number of input array channels does not match " +
"CNN layer configuration"
+ " (data input channels = " + dataInCh
+ df
+ Arrays.toString(input.shape()) + "; expected" + " input channels = " + inWeightChannels + ") "
+ layerId());
}
int[] kernel = layerConfig.getKernelSize();
int[] dilation = layerConfig.getDilation();
int[] strides = layerConfig.getStride();
int[] pad;
int[] outSize;
if (mode == ConvolutionMode.Same) {
outSize = Convolution3DUtils.get3DOutputSize(
input, kernel, strides, null, convolutionMode, dilation, isNCDHW);
int[] inSize = new int[]{inD, inH, inW};
pad = Convolution3DUtils.get3DSameModeTopLeftPadding(outSize,
inSize, kernel, strides, dilation);
} else {
pad = layerConfig.getPadding();
outSize = Convolution3DUtils.get3DOutputSize(input, kernel, strides, pad, convolutionMode, dilation, isNCDHW);
}
int outD = outSize[0];
int outH = outSize[1];
int outW = outSize[2];
INDArray output = workspaceMgr.createUninitialized(ArrayType.ACTIVATIONS, weights.dataType(),miniBatch*outWeightChannels*outD*outH*outW);
if (isNCDHW)
output = output.reshape('c', miniBatch, outWeightChannels, outD, outH, outW);
else
output = output.reshape('c', miniBatch, outD, outH, outW, outWeightChannels);
int[] intArgs = new int[]{
kernel[0], kernel[1], kernel[2],
strides[0], strides[1], strides[2],
pad[0], pad[1], pad[2],
dilation[0], dilation[1], dilation[2],
mode == ConvolutionMode.Same ? 1 : 0,
isNCDHW ? 0 : 1
};
//DL4J conv3d weights: val weightsShape = new long[]{outputDepth, inputDepth, kernel[0], kernel[1], kernel[2]};
//libnd4j conv3d weights: [kD, kH, kW, iC, oC]
weights = weights.permute(2, 3, 4, 1, 0);
INDArray[] inputs;
if (layerConfig.hasBias()) {
INDArray bias = getParamWithNoise(Convolution3DParamInitializer.BIAS_KEY, training, workspaceMgr);
inputs = new INDArray[]{input, weights, bias};
} else {
inputs = new INDArray[]{input, weights};
}
CustomOp op = DynamicCustomOp.builder("conv3dnew")
.addInputs(inputs)
.addIntegerArguments(intArgs)
.addOutputs(output)
.callInplace(false)
.build();
Nd4j.getExecutioner().exec(op);
return new Pair<>(output, null);
}
}
