Please wait. This can take some minutes ...
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.nd4j.linalg.api.ops.impl.layers.convolution.DeConv2DTF 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.nd4j.linalg.api.ops.impl.layers.convolution;
import lombok.Builder;
import lombok.Getter;
import lombok.NoArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import lombok.val;
import org.nd4j.autodiff.samediff.SDVariable;
import org.nd4j.autodiff.samediff.SameDiff;
import org.nd4j.common.base.Preconditions;
import org.nd4j.imports.converters.DifferentialFunctionClassHolder;
import org.nd4j.imports.descriptors.properties.AttributeAdapter;
import org.nd4j.imports.descriptors.properties.PropertyMapping;
import org.nd4j.imports.descriptors.properties.adapters.*;
import org.nd4j.imports.graphmapper.tf.TFGraphMapper;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.api.ops.impl.layers.convolution.config.Conv2DConfig;
import org.nd4j.linalg.api.ops.impl.layers.convolution.config.DeConv2DConfig;
import org.nd4j.common.util.ArrayUtil;
import org.tensorflow.framework.AttrValue;
import org.tensorflow.framework.GraphDef;
import org.tensorflow.framework.NodeDef;
import java.lang.reflect.Field;
import java.util.*;
@Slf4j
@Getter
@NoArgsConstructor
public class DeConv2DTF extends DynamicCustomOp {
protected DeConv2DConfig config;
@Builder(builderMethodName = "sameDiffBuilder")
public DeConv2DTF(SameDiff sameDiff,
SDVariable[] inputs,
DeConv2DConfig config) {
super(sameDiff, inputs);
this.config = config;
addArgs();
}
public DeConv2DTF(INDArray[] inputs, INDArray[] outputs, DeConv2DConfig config){
super(inputs, outputs);
this.config = config;
addArgs();
}
@Override
public long[] iArgs() {
if (iArguments.size() == 0)
addArgs();
return super.iArgs();
}
@Override
public Map propertiesForFunction() {
if(config == null && !iArguments.isEmpty()){
config = DeConv2DConfig.builder()
.kH(iArguments.get(0))
.kW(iArguments.get(1))
.sH(iArguments.get(2))
.sW(iArguments.get(3))
.pH(iArguments.get(4))
.pW(iArguments.get(5))
.dH(iArguments.get(6))
.dW(iArguments.get(7))
.isSameMode(iArguments.get(8) == 1)
.dataFormat(iArguments.get(9) == 1 ? DeConv2DConfig.NHWC : Conv2DConfig.NCHW)
.build();
}
return config.toProperties();
}
private void addArgs() {
addIArgument(config.getKH());
addIArgument(config.getKW());
addIArgument(config.getSH());
addIArgument(config.getSW());
addIArgument(config.getPH());
addIArgument(config.getPW());
addIArgument(config.getDH());
addIArgument(config.getDW());
addIArgument(ArrayUtil.fromBoolean(config.isSameMode()));
addIArgument(config.getDataFormat().equalsIgnoreCase(DeConv2DConfig.NCHW) ? 0 : 1);
}
@Override
public boolean isConfigProperties() {
return true;
}
@Override
public String configFieldName() {
return "config";
}
@Override
public Object getValue(Field property) {
if (config == null) {
config = DeConv2DConfig.builder().build();
}
return config.getValue(property);
}
@Override
public Map> mappingsForFunction() {
Map> ret = new HashMap<>();
Map map = new HashMap<>();
val strideMapping = PropertyMapping.builder()
.tfAttrName("strides")
.onnxAttrName("strides")
.propertyNames(new String[]{"sH", "sW"})
.build();
val kernelMapping = PropertyMapping.builder()
.propertyNames(new String[]{"kH", "kW"})
.tfInputPosition(1)
.onnxAttrName("kernel_shape")
.build();
val dilationMapping = PropertyMapping.builder()
.onnxAttrName("dilations")
.propertyNames(new String[]{"dW", "dH"})
.tfAttrName("rates")
.build();
val sameMode = PropertyMapping.builder()
.onnxAttrName("auto_pad")
.propertyNames(new String[]{"isSameMode"})
.tfAttrName("padding")
.build();
val dataFormat = PropertyMapping.builder()
.onnxAttrName("data_format")
.tfAttrName("data_format")
.propertyNames(new String[]{"dataFormat"})
.build();
map.put("sW", strideMapping);
map.put("sH", strideMapping);
map.put("kH", kernelMapping);
map.put("kW", kernelMapping);
map.put("dW", dilationMapping);
map.put("dH", dilationMapping);
map.put("isSameMode", sameMode);
map.put("dataFormat", dataFormat);
ret.put(tensorflowName(), map);
return ret;
}
@Override
public Map> attributeAdaptersForFunction() {
Map> ret = new HashMap<>();
Map tfMappings = new LinkedHashMap<>();
val fields = DifferentialFunctionClassHolder.getInstance().getFieldsForFunction(this);
//TF uses [kH, kW, outC, inC] always for weights
tfMappings.put("kH", new NDArrayShapeAdapter(0));
tfMappings.put("kW", new NDArrayShapeAdapter(1));
// tfMappings.put("sH", new IntArrayIntIndexAdpater(1));
// tfMappings.put("sW", new IntArrayIntIndexAdpater(2));
tfMappings.put("sH", new ConditionalFieldValueIntIndexArrayAdapter("NCHW", 2, 1, fields.get("dataFormat")));
tfMappings.put("sW", new ConditionalFieldValueIntIndexArrayAdapter("NCHW", 3, 2, fields.get("dataFormat")));
tfMappings.put("isSameMode", new StringEqualsAdapter("SAME"));
tfMappings.put("isNHWC", new StringEqualsAdapter("NHWC"));
Map onnxMappings = new HashMap<>();
onnxMappings.put("kH", new SizeThresholdIntArrayIntIndexAdapter(0, 2, 0));
onnxMappings.put("kW", new SizeThresholdIntArrayIntIndexAdapter(1, 2, 0));
onnxMappings.put("dH", new SizeThresholdIntArrayIntIndexAdapter(0, 2, 0));
onnxMappings.put("dW", new SizeThresholdIntArrayIntIndexAdapter(1, 2, 0));
onnxMappings.put("sH", new SizeThresholdIntArrayIntIndexAdapter(0, 2, 0));
onnxMappings.put("sW", new SizeThresholdIntArrayIntIndexAdapter(1, 2, 0));
onnxMappings.put("isSameMode", new StringEqualsAdapter("SAME"));
onnxMappings.put("isNHWC", new StringEqualsAdapter("NHWC"));
ret.put(tensorflowName(), tfMappings);
return ret;
}
@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map attributesForNode, GraphDef graph) {
TFGraphMapper.initFunctionFromProperties(nodeDef.getOp(), this, attributesForNode, nodeDef, graph);
addArgs();
}
@Override
public String opName() {
return "deconv2d_tf";
}
@Override
public String onnxName() {
return "ConvTranspose-Absent";
}
@Override
public String tensorflowName() {
return "Conv2DBackpropInput";
}
@Override
public List doDiff(List f1) {
throw new UnsupportedOperationException("To be implemented yet");
}
@Override
public List calculateOutputDataTypes(List inputDataTypes){ //inShape, weights, input
int n = args().length;
Preconditions.checkState(inputDataTypes != null && inputDataTypes.size() == n, "Expected %s input data types for %s, got %s", n, getClass(), inputDataTypes);
if(!dArguments.isEmpty()) {
return Arrays.asList(dArguments.get(0));
}
return Collections.singletonList(inputDataTypes.get(2));
}
}