org.tensorflow.framework.NodeDef#getAttrOrThrow

Source File: StridedSlice.java From deeplearning4j with Apache License 2.0

6 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val inputBegin = nodeDef.getInput(1);
    val inputEnd = nodeDef.getInput(2);
    val inputStrides = nodeDef.getInput(3);

    // bit masks for this slice
    val bm = nodeDef.getAttrOrThrow("begin_mask");
    val xm = nodeDef.getAttrOrThrow("ellipsis_mask");
    val em = nodeDef.getAttrOrThrow("end_mask");
    val nm = nodeDef.getAttrOrThrow("new_axis_mask");
    val sm = nodeDef.getAttrOrThrow("shrink_axis_mask");

    beginMask = (int)bm.getI();
    ellipsisMask = (int) xm.getI();
    endMask = (int) em.getI();
    newAxisMask = (int) nm.getI();
    shrinkAxisMask = (int) sm.getI();

    addIArgument(beginMask);
    addIArgument(ellipsisMask);
    addIArgument(endMask);
    addIArgument(newAxisMask);
    addIArgument(shrinkAxisMask);
}

Source File: LocalResponseNormalization.java From nd4j with Apache License 2.0

6 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {

    val aAlpha = nodeDef.getAttrOrThrow("alpha");
    val aBeta = nodeDef.getAttrOrThrow("beta");
    val aBias = nodeDef.getAttrOrThrow("bias");
    val aDepth = nodeDef.getAttrOrThrow("depth_radius");

    val alpha = aAlpha.getF();
    val beta = aBeta.getF();
    val bias = aBias.getF();
    val depth = aDepth.getF();

    LocalResponseNormalizationConfig localResponseNormalizationConfig = LocalResponseNormalizationConfig.builder()
            .alpha(alpha)
            .beta(beta)
            .bias(bias)
            .depth((int) depth)
            .build();
    this.config = localResponseNormalizationConfig;
    addArgs();
}

Source File: LocalResponseNormalization.java From deeplearning4j with Apache License 2.0

6 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {

    val aAlpha = nodeDef.getAttrOrThrow("alpha");
    val aBeta = nodeDef.getAttrOrThrow("beta");
    val aBias = nodeDef.getAttrOrThrow("bias");
    val aDepth = nodeDef.getAttrOrThrow("depth_radius");

    double alpha = aAlpha.getF();
    double beta = aBeta.getF();
    double bias = aBias.getF();
    int depth = (int)aDepth.getI();

    LocalResponseNormalizationConfig localResponseNormalizationConfig = LocalResponseNormalizationConfig.builder()
            .alpha(alpha)
            .beta(beta)
            .bias(bias)
            .depth((int) depth)
            .build();
    this.config = localResponseNormalizationConfig;
    addArgs();
}

Source File: Unstack.java From deeplearning4j with Apache License 2.0

5 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val attrAxis = nodeDef.getAttrOrThrow("axis");
    int axis = (int) attrAxis.getI();
    this.jaxis = axis;
    val attrNum = nodeDef.getAttrOrDefault("num", null);
    if(attrNum != null){
        this.num = (int) attrNum.getI();
    }
    addArgs();
}

Source File: Unstack.java From nd4j with Apache License 2.0

5 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val attrAxis = nodeDef.getAttrOrThrow("axis");
    int axis = (int) attrAxis.getI();
    this.axis = axis;
    addArgs();
}

Source File: Pooling2D.java From deeplearning4j with Apache License 2.0

5 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();
    val sH = tfStrides.get(1);
    val sW = tfStrides.get(2);

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    val kH = tfKernels.get(1);
    val kW = tfKernels.get(2);

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"","");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    if (!isSameMode)
        log.debug("Mode: {}", paddingMode);

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sH(sH.intValue())
            .sW(sW.intValue())
            .type(null)
            .isSameMode(isSameMode)
            .kH(kH.intValue())
            .kW(kW.intValue())
            .pH(padding.get(0).intValue())
            .pW(padding.get(1).intValue())
            .build();
    this.config = pooling2DConfig;
    addArgs();
    log.debug("Pooling: k: [{},{}]; s: [{}, {}], padding: {}", kH, kW, sH, sW, aPadding);


}

Source File: MaxPooling2D.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    int sH = 0;
    int sW = 0;

    int pH = 0;
    int pW = 0;

    int kH = 0;
    int kW = 0;

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"", "");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    if (data_format.equalsIgnoreCase("nhwc")) {
        sH = tfStrides.get(1).intValue();
        sW = tfStrides.get(2).intValue();

        kH = tfKernels.get(1).intValue();
        kW = tfKernels.get(2).intValue();

        pH = padding.size() > 0 ? padding.get(1).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(2).intValue() : 0;
    } else {
        sH = tfStrides.get(2).intValue();
        sW = tfStrides.get(3).intValue();

        kH = tfKernels.get(2).intValue();
        kW = tfKernels.get(3).intValue();

        pH = padding.size() > 0 ? padding.get(2).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(3).intValue() : 0;
    }

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sH(sH)
            .sW(sW)
            .type(Pooling2D.Pooling2DType.MAX)
            .isSameMode(isSameMode)
            .kH(kH)
            .kW(kW)
            .pH(pH)
            .pW(pW)
            .isNHWC(data_format.equalsIgnoreCase("nhwc"))
            .extra(1.0) // averaging only for non-padded values
            .build();
    this.config = pooling2DConfig;
    addArgs();
}

Source File: AvgPooling2D.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    int sH = 0;
    int sW = 0;

    int pH = 0;
    int pW = 0;

    int kH = 0;
    int kW = 0;

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"", "");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    if (data_format.equalsIgnoreCase("nhwc")) {
        sH = tfStrides.get(1).intValue();
        sW = tfStrides.get(2).intValue();

        kH = tfKernels.get(1).intValue();
        kW = tfKernels.get(2).intValue();

        pH = padding.size() > 0 ? padding.get(1).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(2).intValue() : 0;
    } else {
        sH = tfStrides.get(2).intValue();
        sW = tfStrides.get(3).intValue();

        kH = tfKernels.get(2).intValue();
        kW = tfKernels.get(3).intValue();

        pH = padding.size() > 0 ? padding.get(2).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(3).intValue() : 0;
    }

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sH(sH)
            .sW(sW)
            .type(Pooling2D.Pooling2DType.AVG)
            .isSameMode(isSameMode)
            .kH(kH)
            .kW(kW)
            .pH(pH)
            .pW(pW)
            .isNHWC(data_format.equalsIgnoreCase("nhwc"))
            .extra(0.0) // averaging only for non-padded values
            .build();
    this.config = pooling2DConfig;
    addArgs();
    //log.debug("Pooling: k: [{},{}]; s: [{}, {}], padding: {}", kH, kW, sH, sW, aPadding);
}

Source File: Pooling3D.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    List<Long> tfStrides = aStrides.getList().getIList();
    val aKernels = nodeDef.getAttrOrThrow("ksize");
    List<Long> tfKernels = aKernels.getList().getIList();
    val aPadding = nodeDef.getAttrOrThrow("padding");
    List<Long> tfPadding = aPadding.getList().getIList();

    String paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"", "");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "ndhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    //Order: depth, height, width
    //TF doesn't have dilation, it seems?
    int[] strides = new int[3];
    int[] padding = new int[3];
    int[] kernel = new int[3];
    for( int i=0; i<3; i++ ) {
        //TF values here have 5 values: minibatch and Channels at positions 0 and 4, which are almost always 1
        strides[i] = tfStrides.get(i+1).intValue();
        if(tfPadding != null && tfPadding.size() > 0) {
            //Empty for SAME mode
            padding[i] = tfPadding.get(i + 1).intValue();
        }
        kernel[i] = tfKernels.get(i+1).intValue();
    }

    Pooling3DType type;
    String name = nodeDef.getOp().toLowerCase();
    if(name.startsWith("max")){
        type = Pooling3DType.MAX;
    } else if(name.startsWith("av")){
        type = Pooling3DType.AVG;
    } else {
        throw new IllegalStateException("Unknown or not supported pooling type: " + name);
    }

    Pooling3DConfig conf = Pooling3DConfig.builder()
            .sD(strides[0]).sH(strides[1]).sW(strides[2])
            .pD(padding[0]).pH(padding[1]).pW(padding[2])
            .kD(kernel[0]).kH(kernel[1]).kW(kernel[2])
            .type(type)
            .isSameMode(isSameMode)
            .isNCDHW(data_format.equalsIgnoreCase("ncdhw"))
            .build();
    this.config = conf;
    addArgs();
}

Source File: MaxPoolWithArgmax.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    int sH = 0;
    int sW = 0;

    int pH = 0;
    int pW = 0;

    int kH = 0;
    int kW = 0;

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"", "");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    if (data_format.equalsIgnoreCase("nhwc")) {
        sH = tfStrides.get(1).intValue();
        sW = tfStrides.get(2).intValue();

        kH = tfKernels.get(1).intValue();
        kW = tfKernels.get(2).intValue();

        pH = padding.size() > 0 ? padding.get(1).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(2).intValue() : 0;
    } else {
        sH = tfStrides.get(2).intValue();
        sW = tfStrides.get(3).intValue();

        kH = tfKernels.get(2).intValue();
        kW = tfKernels.get(3).intValue();

        pH = padding.size() > 0 ? padding.get(2).intValue() : 0;
        pW = padding.size() > 0 ? padding.get(3).intValue() : 0;
    }

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sH(sH)
            .sW(sW)
            .type(Pooling2D.Pooling2DType.MAX)
            .isSameMode(isSameMode)
            .kH(kH)
            .kW(kW)
            .pH(pH)
            .pW(pW)
            .isNHWC(data_format.equalsIgnoreCase("nhwc"))
            .extra(1.0) // averaging only for non-padded values
            .build();
    this.config = pooling2DConfig;
    addArgs();
    if(attributesForNode.containsKey("argmax")) {
        outputType = TFGraphMapper.convertType(attributesForNode.get("argmax").getType());
    } else {
        outputType = DataType.LONG;
    }
}

Source File: DeConv2D.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();
    long sH = 1;
    long sW = 1;
    long kH = 1;
    long kW = 1;

    val aPadding = nodeDef.getAttrOrDefault("padding", null);

    val paddingMode = aPadding.getS().toStringUtf8();

    val args = args();
    INDArray arr = sameDiff.getVariable(args[1].name()).getArr();
    if (arr == null) {
        arr = TFGraphMapper.getNDArrayFromTensor(nodeDef);
        // TODO: arguable. it might be easier to permute weights once
        //arr = (arr.permute(3, 2, 0, 1).dup('c'));
        val varForOp = initWith.getVariable(args[1].name());
        if (arr != null)
            initWith.associateArrayWithVariable(arr, varForOp);


    }

    String dataFormat = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");
        dataFormat = attr.getS().toStringUtf8().toLowerCase();
    }

    if (dataFormat.equalsIgnoreCase(DeConv2DConfig.NCHW)) {
        sH = tfStrides.get(2).longValue();
        sW = tfStrides.get(3).longValue();

        kH = arr.size(2);
        kW = arr.size(3);
    } else {
        sH = tfStrides.get(1).longValue();
        sW = tfStrides.get(2).longValue();

        kH = arr.size(0);
        kW = arr.size(1);
    }


    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");
    DeConv2DConfig conv2DConfig = DeConv2DConfig.builder()
            .kH(kH)
            .kW(kW)
            .sH(sW)
            .sW(sH)
            .isSameMode(isSameMode)
            .dataFormat(dataFormat.equalsIgnoreCase(DeConv2DConfig.NHWC) ? DeConv2DConfig.NHWC : DeConv2DConfig.NCHW)
            .build();
    this.config = conv2DConfig;

    addArgs();


}

Source File: MaxPooling2D.java From nd4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    int sY = 0;
    int sX = 0;

    int ph = 0;
    int pw = 0;

    int kY = 0;
    int kX = 0;

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"", "");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    if (data_format.equalsIgnoreCase("nhwc")) {
        sY = tfStrides.get(1).intValue();
        sX = tfStrides.get(2).intValue();

        kY = tfKernels.get(1).intValue();
        kX = tfKernels.get(2).intValue();

        ph = padding.size() > 0 ? padding.get(1).intValue() : 0;
        pw = padding.size() > 0 ? padding.get(2).intValue() : 0;
    } else {
        sY = tfStrides.get(2).intValue();
        sX = tfStrides.get(3).intValue();

        kY = tfKernels.get(2).intValue();
        kX = tfKernels.get(3).intValue();

        ph = padding.size() > 0 ? padding.get(2).intValue() : 0;
        pw = padding.size() > 0 ? padding.get(3).intValue() : 0;
    }

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sy(sY)
            .sx(sX)
            .type(Pooling2D.Pooling2DType.MAX)
            .isSameMode(isSameMode)
            .kh(kY)
            .kw(kX)
            .ph(ph)
            .pw(pw)
            .virtualWidth(1)
            .virtualHeight(1)
            .isNHWC(data_format.equalsIgnoreCase("nhwc"))
            .extra(1.0) // averaging only for non-padded values
            .build();
    this.config = pooling2DConfig;
    addArgs();
    log.debug("Pooling: k: [{},{}]; s: [{}, {}], padding: {}", kY, kX, sY, sX, aPadding);


}

Source File: DeConv3DTF.java From deeplearning4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {

    val aStrides = nodeDef.getAttrOrThrow("strides");
    val aDilations = nodeDef.getAttrOrDefault("dilations", null);
    val tfStrides = aStrides.getList().getIList();
    val tfDilation = aDilations == null ? null : aDilations.getList().getIList();
    int sD, sH, sW, dD, dH, dW;

    val aPadding = nodeDef.getAttrOrDefault("padding", null);
    String paddingMode = aPadding.getS().toStringUtf8();

    String dataFormat = DeConv3DConfig.NDHWC;
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");
        dataFormat = attr.getS().toStringUtf8().toLowerCase();
    }

    if (dataFormat.equalsIgnoreCase(DeConv3DConfig.NCDHW)) {
        sD = tfStrides.get(2).intValue();
        sH = tfStrides.get(3).intValue();
        sW = tfStrides.get(4).intValue();


        dD = tfDilation == null ? 1 : tfDilation.get(2).intValue();
        dH = tfDilation == null ? 1 : tfDilation.get(3).intValue();
        dW = tfDilation == null ? 1 : tfDilation.get(4).intValue();
    } else {
        sD = tfStrides.get(1).intValue();
        sH = tfStrides.get(2).intValue();
        sW = tfStrides.get(3).intValue();

        dD = tfDilation == null ? 1 : tfDilation.get(1).intValue();
        dH = tfDilation == null ? 1 : tfDilation.get(2).intValue();
        dW = tfDilation == null ? 1 : tfDilation.get(3).intValue();
    }


    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");
    DeConv3DConfig conv3DConfig = DeConv3DConfig.builder()
            .kD(-1)
            .kH(-1)
            .kW(-1)
            .sD(sD)
            .sH(sW)
            .sW(sH)
            .dD(dD)
            .dH(dH)
            .dW(dW)
            .isSameMode(isSameMode)
            .dataFormat(dataFormat.equalsIgnoreCase(DeConv3DConfig.NCDHW) ? DeConv3DConfig.NCDHW : DeConv3DConfig.NDHWC)
            .build();
    this.config = conv3DConfig;

    addArgs();
}

Source File: StridedSlice.java From nd4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val inputBegin = nodeDef.getInput(1);
    val inputEnd = nodeDef.getInput(2);
    val inputStrides = nodeDef.getInput(3);

    NodeDef beginNode = null;
    NodeDef endNode = null;
    NodeDef strides = null;

    for(int i = 0; i < graph.getNodeCount(); i++) {
        if(graph.getNode(i).getName().equals(inputBegin)) {
            beginNode = graph.getNode(i);
        }
        if(graph.getNode(i).getName().equals(inputEnd)) {
            endNode = graph.getNode(i);
        }
        if(graph.getNode(i).getName().equals(inputStrides)) {
            strides = graph.getNode(i);
        }
    }


    // bit masks for this slice
    val bm = nodeDef.getAttrOrThrow("begin_mask");
    val xm = nodeDef.getAttrOrThrow("ellipsis_mask");
    val em = nodeDef.getAttrOrThrow("end_mask");
    val nm = nodeDef.getAttrOrThrow("new_axis_mask");
    val sm = nodeDef.getAttrOrThrow("shrink_axis_mask");

    addIArgument((int) bm.getI());
    addIArgument((int) xm.getI());
    addIArgument((int) em.getI());

    addIArgument((int) nm.getI());
    addIArgument((int) sm.getI());

    val beginArr = TFGraphMapper.getInstance().getNDArrayFromTensor("value",beginNode,graph);
    val endArr = TFGraphMapper.getInstance().getNDArrayFromTensor("value",endNode,graph);
    val stridesArr = TFGraphMapper.getInstance().getNDArrayFromTensor("value",strides,graph);

    if (beginArr != null && endArr != null && stridesArr != null) {

        for (int e = 0; e < beginArr.length(); e++)
            addIArgument(beginArr.getInt(e));

        for (int e = 0; e <  endArr.length(); e++)
            addIArgument(endArr.getInt(e));

        for (int e = 0; e < stridesArr.length(); e++)
            addIArgument(stridesArr.getInt(e));
    } else {
        // do nothing
    }



}

Source File: AvgPooling2D.java From nd4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    int sY = 0;
    int sX = 0;

    int ph = 0;
    int pw = 0;

    int kY = 0;
    int kX = 0;

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"","");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    String data_format = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");

        data_format = attr.getS().toStringUtf8().toLowerCase();
    }

    if (data_format.equalsIgnoreCase("nhwc")) {
        sY = tfStrides.get(1).intValue();
        sX = tfStrides.get(2).intValue();

        kY = tfKernels.get(1).intValue();
        kX = tfKernels.get(2).intValue();

        ph = padding.size() > 0 ? padding.get(1).intValue() : 0;
        pw = padding.size() > 0 ? padding.get(2).intValue() : 0;
    } else {
        sY = tfStrides.get(2).intValue();
        sX = tfStrides.get(3).intValue();

        kY = tfKernels.get(2).intValue();
        kX = tfKernels.get(3).intValue();

        ph = padding.size() > 0 ? padding.get(2).intValue() : 0;
        pw = padding.size() > 0 ? padding.get(3).intValue() : 0;
    }

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sy(sY)
            .sx(sX)
            .type(Pooling2D.Pooling2DType.AVG)
            .isSameMode(isSameMode)
            .kh(kY)
            .kw(kX)
            .ph(ph)
            .pw(pw)
            .virtualWidth(1)
            .virtualHeight(1)
            .isNHWC(data_format.equalsIgnoreCase("nhwc"))
            .extra(0.0) // averaging only for non-padded values
            .build();
    this.config = pooling2DConfig;
    addArgs();
    log.debug("Pooling: k: [{},{}]; s: [{}, {}], padding: {}", kY, kX, sY, sX, aPadding);


}

Source File: Pooling2D.java From nd4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();
    val sY = tfStrides.get(1);
    val sX = tfStrides.get(2);

    val aKernels = nodeDef.getAttrOrThrow("ksize");
    val tfKernels = aKernels.getList().getIList();

    val kY = tfKernels.get(1);
    val kX = tfKernels.get(2);

    val aPadding = nodeDef.getAttrOrThrow("padding");
    val padding = aPadding.getList().getIList();

    val paddingMode = aPadding.getS().toStringUtf8().replaceAll("\"","");

    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");

    if (!isSameMode)
        log.debug("Mode: {}", paddingMode);

    Pooling2DConfig pooling2DConfig = Pooling2DConfig.builder()
            .sy(sY.intValue())
            .sx(sX.intValue())
            .type(null)
            .isSameMode(isSameMode)
            .kh(kY.intValue())
            .kw(kX.intValue())
            .ph(padding.get(0).intValue())
            .pw(padding.get(1).intValue())
            .virtualWidth(1)
            .virtualHeight(1)
            .build();
    this.config = pooling2DConfig;
    addArgs();
    log.debug("Pooling: k: [{},{}]; s: [{}, {}], padding: {}", kY, kX, sY, sX, aPadding);


}

Source File: DeConv2D.java From nd4j with Apache License 2.0

4 votes

@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map<String, AttrValue> attributesForNode, GraphDef graph) {
    val aStrides = nodeDef.getAttrOrThrow("strides");
    val tfStrides = aStrides.getList().getIList();
    int sY = 1;
    int sX = 1;
    int kY = 1;
    int kX = 1;

    val aPadding = nodeDef.getAttrOrDefault("padding", null);

    val paddingMode = aPadding.getS().toStringUtf8();

    val args = args();
    INDArray arr = sameDiff.getVariable(args[1].getVarName()).getArr();
    if (arr == null) {
        arr = TFGraphMapper.getInstance().getNDArrayFromTensor(nodeDef.getInput(0), nodeDef, graph);
        // TODO: arguable. it might be easier to permute weights once
        //arr = (arr.permute(3, 2, 0, 1).dup('c'));
        val varForOp = initWith.getVariable(args[1].getVarName());
        if (arr != null)
            initWith.associateArrayWithVariable(arr, varForOp);


    }

    String dataFormat = "nhwc";
    if (nodeDef.containsAttr("data_format")) {
        val attr = nodeDef.getAttrOrThrow("data_format");
        dataFormat = attr.getS().toStringUtf8().toLowerCase();
    }

    // FIXME: int cast


    if (dataFormat.equalsIgnoreCase("nchw")) {
        sY = tfStrides.get(2).intValue();
        sX = tfStrides.get(3).intValue();

        kY = (int) arr.size(2);
        kX = (int) arr.size(3);
    } else {
        sY = tfStrides.get(1).intValue();
        sX = tfStrides.get(2).intValue();

        kY = (int) arr.size(0);
        kX = (int) arr.size(1);
    }


    boolean isSameMode = paddingMode.equalsIgnoreCase("SAME");
    DeConv2DConfig conv2DConfig = DeConv2DConfig.builder()
            .kY(kY)
            .kX(kX)
            .sX(sX)
            .sY(sY)
            .isSameMode(isSameMode)
            //c++ check checks for nchw
            .isNHWC(dataFormat.equalsIgnoreCase("nhwc"))
            .build();
    this.config = conv2DConfig;

    addArgs();


}

Java Code Examples for org.tensorflow.framework.NodeDef#getAttrOrThrow()