com.simiacryptus.mindseye.art.models.VGG19_HDF5 Maven / Gradle / Ivy
/*
* 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.art.models;
import com.simiacryptus.mindseye.art.util.Hdf5Archive;
import com.simiacryptus.mindseye.lang.Layer;
import com.simiacryptus.mindseye.lang.Tensor;
import com.simiacryptus.mindseye.layers.Explodable;
import com.simiacryptus.mindseye.layers.cudnn.*;
import com.simiacryptus.mindseye.layers.cudnn.conv.ConvolutionLayer;
import com.simiacryptus.mindseye.network.DAGNetwork;
import com.simiacryptus.mindseye.network.PipelineNetwork;
import com.simiacryptus.mindseye.test.TestUtil;
import com.simiacryptus.util.Util;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import javax.annotation.Nonnull;
import java.io.File;
import java.io.IOException;
import java.security.KeyManagementException;
import java.security.NoSuchAlgorithmException;
class VGG19_HDF5 {
public static final Logger log = LoggerFactory.getLogger(VGG19_HDF5.class);
public final Hdf5Archive hdf5;
@Nonnull
int[] convolutionOrder = {3, 2, 0, 1};
@Nonnull
int[] fullyconnectedOrder = {1, 0};
public VGG19_HDF5(Hdf5Archive hdf5) {
this.hdf5 = hdf5;
}
@Nonnull
protected static Layer add(@Nonnull Layer layer, @Nonnull PipelineNetwork model) {
if (layer instanceof Explodable) {
Layer explode = ((Explodable) layer).explode();
try {
if (explode instanceof DAGNetwork) {
log.info(String.format(
"Exploded %s to %s (%s nodes)",
layer.getName(),
explode.getClass().getSimpleName(),
((DAGNetwork) explode).getNodes().size()
));
} else {
log.info(String.format("Exploded %s to %s (%s nodes)", layer.getName(), explode.getClass().getSimpleName(), explode.getName()));
}
return add(explode, model);
} finally {
layer.freeRef();
}
} else {
model.wrap(layer).freeRef();
return layer;
}
}
public static VGG19_HDF5 fromHDF5() {
try {
return fromHDF5(Util.cacheFile(TestUtil.S3_ROOT.resolve("vgg19_weights.h5")));
} catch (IOException | KeyManagementException | NoSuchAlgorithmException e) {
throw new RuntimeException(e);
}
}
public static VGG19_HDF5 fromHDF5(final File hdf) {
try {
return new VGG19_HDF5(new Hdf5Archive(hdf));
} catch (@Nonnull final RuntimeException e) {
throw e;
} catch (Throwable e) {
throw new RuntimeException(e);
}
}
public void phase0(PipelineNetwork pipeline) {
add(new ImgBandBiasLayer(3).setAndFree(new Tensor(-103.939, -116.779, -123.68)), pipeline);
addConvolutionLayer(3, 3, 64, ActivationLayer.Mode.RELU, "layer_1", pipeline);
}
public void phase1a(PipelineNetwork pipeline) {
addConvolutionLayer(3, 64, 64, ActivationLayer.Mode.RELU, "layer_3", pipeline);
}
public void phase1b2(PipelineNetwork pipeline) {
addConvolutionLayer(3, 128, 128, ActivationLayer.Mode.RELU, "layer_8", pipeline);
}
public void phase1b1(PipelineNetwork pipeline) {
addPoolingLayer(2, pipeline);
addConvolutionLayer(3, 64, 128, ActivationLayer.Mode.RELU, "layer_6", pipeline);
}
public void phase1c4(PipelineNetwork pipeline) {
addConvolutionLayer(3, 256, 256, ActivationLayer.Mode.RELU, "layer_17", pipeline);
}
public void phase1c3(PipelineNetwork pipeline) {
addConvolutionLayer(3, 256, 256, ActivationLayer.Mode.RELU, "layer_15", pipeline);
}
public void phase1c2(PipelineNetwork pipeline) {
addConvolutionLayer(3, 256, 256, ActivationLayer.Mode.RELU, "layer_13", pipeline);
}
public void phase1c1(PipelineNetwork pipeline) {
addPoolingLayer(2, pipeline);
addConvolutionLayer(3, 128, 256, ActivationLayer.Mode.RELU, "layer_11", pipeline);
}
public void phase1d4(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_26", pipeline);
}
public void phase1d3(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_24", pipeline);
}
public void phase1d2(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_22", pipeline);
}
public void phase1d1(PipelineNetwork pipeline) {
addPoolingLayer(2, pipeline);
addConvolutionLayer(3, 256, 512, ActivationLayer.Mode.RELU, "layer_20", pipeline);
}
public void phase1e4(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_35", pipeline);
}
public void phase1e3(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_33", pipeline);
}
public void phase1e2(PipelineNetwork pipeline) {
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_31", pipeline);
}
public void phase1e1(PipelineNetwork pipeline) {
addPoolingLayer(2, pipeline);
addConvolutionLayer(3, 512, 512, ActivationLayer.Mode.RELU, "layer_29", pipeline);
}
public void phase2(PipelineNetwork pipeline) {
// model.add(MaxPooling2D((2,2), strides=(2,2)))
addPoolingLayer(2, pipeline);
if (true) {
add(new ImgModulusPaddingLayer(7, 7), pipeline);
} else {
add(new ImgModulusPaddingLayer(-7, -7), pipeline);
}
add(new ConvolutionLayer(7, 7, 512, 4096)
.setStrideXY(1, 1)
.setPaddingXY(0, 0)
.setAndFree(hdf5.readDataSet("param_0", "layer_38")
.reshapeCastAndFree(7, 7, 512, 4096).permuteDimensionsAndFree(0, 1, 3, 2)
), pipeline);
add(new ImgBandBiasLayer(4096)
.setAndFree((hdf5.readDataSet("param_1", "layer_38"))), pipeline);
add(new ActivationLayer(ActivationLayer.Mode.RELU), pipeline);
}
public void phase3a(PipelineNetwork pipeline) {
add(new ConvolutionLayer(1, 1, 4096, 4096)
.setPaddingXY(0, 0)
.setAndFree(hdf5.readDataSet("param_0", "layer_40")
.permuteDimensionsAndFree(fullyconnectedOrder)), pipeline);
add(new ImgBandBiasLayer(4096)
.setAndFree((hdf5.readDataSet("param_1", "layer_40"))), pipeline);
add(new ActivationLayer(ActivationLayer.Mode.RELU), pipeline);
}
public void addPoolingLayer(final int size, PipelineNetwork pipeline) {
if (true) {
add(new ImgModulusPaddingLayer(size, size), pipeline);
} else {
Layer layer = new ImgModulusPaddingLayer(-size, -size);
add(layer, pipeline);
}
add(new PoolingLayer()
.setMode(PoolingLayer.PoolingMode.Max)
.setWindowXY(size, size)
.setStrideXY(size, size), pipeline);
}
public void addConvolutionLayer(final int radius, final int inputBands, final int outputBands, final ActivationLayer.Mode activationMode, final String hdf_group, PipelineNetwork pipeline) {
add(new ConvolutionLayer(radius, radius, inputBands, outputBands)
.setPaddingXY(0, 0)
.setAndFree(hdf5.readDataSet("param_0", hdf_group)
.permuteDimensionsAndFree(convolutionOrder)), pipeline);
add(new ImgBandBiasLayer(outputBands)
.setAndFree((hdf5.readDataSet("param_1", hdf_group))), pipeline);
add(new ActivationLayer(activationMode), pipeline);
}
public void phase3b(PipelineNetwork pipeline) {
add(new ConvolutionLayer(1, 1, 4096, 1000)
.setPaddingXY(0, 0)
.setAndFree(hdf5.readDataSet("param_0", "layer_42")
.permuteDimensionsAndFree(fullyconnectedOrder)), pipeline);
add(new ImgBandBiasLayer(1000)
.setAndFree((hdf5.readDataSet("param_1", "layer_42"))), pipeline);
add(new SoftmaxActivationLayer()
.setAlgorithm(SoftmaxActivationLayer.SoftmaxAlgorithm.ACCURATE)
.setMode(SoftmaxActivationLayer.SoftmaxMode.CHANNEL), pipeline);
add(new BandReducerLayer()
.setMode(PoolingLayer.PoolingMode.Max), pipeline);
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy