public TensorShape(TensorShapeProto proto)
{
if (proto.UnknownRank)
{
return;
}
switch (proto.Dim.Count)
{
case 0: shape = new Shape(new int[0]); break;
case 1: shape = Shape.Vector((int)proto.Dim[0].Size); break;
case 2: shape = Shape.Matrix((int)proto.Dim[0].Size, (int)proto.Dim[1].Size); break;
default:
var protodims = proto.Dim;
var len = protodims.Count;
var dims = new int[len];
for (int i = 0; i < len; i++)
{
dims[i] = (int)protodims[i].Size;
}
shape = new Shape(dims); break;
}
}
public static TensorShapeProto as_shape <T>(T[] dims)
{
TensorShapeProto shape = new TensorShapeProto();
for (int i = 0; i < dims.Length; i++)
{
var dim = new TensorShapeProto.Types.Dim();
switch (dims[i])
{
case int n:
dim.Size = n;
break;
case long l:
dim.Size = l;
break;
default:
throw new NotImplementedException("as_shape Not Implemented");
}
// dim.Name = $"dim_{i}";
shape.Dim.Add(dim);
}
return(shape);
}
public TensorShape(TensorShapeProto proto)
{
if (proto.UnknownRank)
{
return;
}
Reshape(proto.Dim.Select(x => (int)x.Size).ToArray());
}
public static TensorShapeProto as_proto(this Shape tshape)
{
TensorShapeProto shape = new TensorShapeProto();
for (int i = 0; i < tshape.ndim; i++)
{
var dim = new TensorShapeProto.Types.Dim();
dim.Size = tshape.dims[i];
//dim.Name = $"dim_{i}";
shape.Dim.Add(dim);
}
return(shape);
}
public static TensorShapeProto as_shape(long[] dims)
{
TensorShapeProto shape = new TensorShapeProto();
for (int i = 0; i < dims.Length; i++)
{
var dim = new TensorShapeProto.Types.Dim();
dim.Size = dims[i];
dim.Name = $"dim_{i}";
shape.Dim.Add(dim);
}
return(shape);
}
public object get_attr(string name)
{
object ret = null;
var fields = new string[] { "s", "i", "f", "b", "type", "shape", "tensor", "func" };
switch (name)
{
case "dtype":
ret = _outputs[0];
break;
case "shape":
ret = new TensorShapeProto();
break;
}
return(ret);
}
public void TestInception()
{
using (Tensor imageTensor = ImageIO.ReadTensorFromImageFile <float>("grace_hopper.jpg", 224, 224, 128.0f, 1.0f))
using (Inception inceptionGraph = new Inception())
{
bool processCompleted = false;
inceptionGraph.OnDownloadCompleted += (sender, e) =>
{
HashSet <string> opNames = new HashSet <string>();
HashSet <string> couldBeInputs = new HashSet <string>();
HashSet <string> couldBeOutputs = new HashSet <string>();
foreach (Operation op in inceptionGraph.Graph)
{
String name = op.Name;
opNames.Add(name);
if (op.NumInputs == 0 && op.OpType.Equals("Placeholder"))
{
couldBeInputs.Add(op.Name);
AttrMetadata dtypeMeta = op.GetAttrMetadata("dtype");
AttrMetadata shapeMeta = op.GetAttrMetadata("shape");
DataType type = op.GetAttrType("dtype");
Int64[] shape = op.GetAttrShape("shape");
Buffer valueBuffer = op.GetAttrValueProto("shape");
Buffer shapeBuffer = op.GetAttrTensorShapeProto("shape");
Tensorflow.TensorShapeProto shapeProto =
Tensorflow.TensorShapeProto.Parser.ParseFrom(shapeBuffer.Data);
}
if (op.OpType.Equals("Const"))
{
AttrMetadata dtypeMeta = op.GetAttrMetadata("dtype");
AttrMetadata valueMeta = op.GetAttrMetadata("value");
using (Tensor valueTensor = op.GetAttrTensor("value"))
{
var dim = valueTensor.Dim;
}
}
if (op.OpType.Equals("Conv2D"))
{
AttrMetadata stridesMeta = op.GetAttrMetadata("strides");
AttrMetadata paddingMeta = op.GetAttrMetadata("padding");
AttrMetadata boolMeta = op.GetAttrMetadata("use_cudnn_on_gpu");
Int64[] strides = op.GetAttrIntList("strides");
bool useCudnn = op.GetAttrBool("use_cudnn_on_gpu");
String padding = op.GetAttrString("padding");
}
foreach (Output output in op.Outputs)
{
int[] shape = inceptionGraph.Graph.GetTensorShape(output);
if (output.NumConsumers == 0)
{
couldBeOutputs.Add(name);
}
}
Buffer buffer = inceptionGraph.Graph.GetOpDef(op.OpType);
Tensorflow.OpDef opDef = Tensorflow.OpDef.Parser.ParseFrom(buffer.Data);
}
using (Buffer versionDef = inceptionGraph.Graph.Versions())
{
int l = versionDef.Length;
}
Inception.RecognitionResult[] results = inceptionGraph.Recognize(imageTensor);
Trace.WriteLine(String.Format("Object is {0} with {1}% probability", results[0].Label, results[0].Probability * 100));
processCompleted = true;
};
inceptionGraph.Init();
while (!processCompleted)
{
Thread.Sleep(1000);
}
}
}
public async Task TestResnet()
{
using (Tensor imageTensor = ImageIO.ReadTensorFromImageFile <float>("surfers.jpg", 224, 224, 0, 1.0f / 255.0f))
using (Resnet resnet = new Resnet())
{
await resnet.Init();
MetaGraphDef metaGraphDef = MetaGraphDef.Parser.ParseFrom(resnet.MetaGraphDefBuffer.Data);
var signatureDef = metaGraphDef.SignatureDef["serving_default"];
var inputNode = signatureDef.Inputs;
var outputNode = signatureDef.Outputs;
HashSet <string> opNames = new HashSet <string>();
HashSet <string> couldBeInputs = new HashSet <string>();
HashSet <string> couldBeOutputs = new HashSet <string>();
foreach (Operation op in resnet.Graph)
{
String name = op.Name;
opNames.Add(name);
if (op.NumInputs == 0 && op.OpType.Equals("Placeholder"))
{
couldBeInputs.Add(op.Name);
AttrMetadata dtypeMeta = op.GetAttrMetadata("dtype");
AttrMetadata shapeMeta = op.GetAttrMetadata("shape");
DataType type = op.GetAttrType("dtype");
Int64[] shape = op.GetAttrShape("shape");
Buffer valueBuffer = op.GetAttrValueProto("shape");
Buffer shapeBuffer = op.GetAttrTensorShapeProto("shape");
Tensorflow.TensorShapeProto shapeProto =
Tensorflow.TensorShapeProto.Parser.ParseFrom(shapeBuffer.Data);
}
if (op.OpType.Equals("Const"))
{
AttrMetadata dtypeMeta = op.GetAttrMetadata("dtype");
AttrMetadata valueMeta = op.GetAttrMetadata("value");
using (Tensor valueTensor = op.GetAttrTensor("value"))
{
var dim = valueTensor.Dim;
}
}
if (op.OpType.Equals("Conv2D"))
{
AttrMetadata stridesMeta = op.GetAttrMetadata("strides");
AttrMetadata paddingMeta = op.GetAttrMetadata("padding");
AttrMetadata boolMeta = op.GetAttrMetadata("use_cudnn_on_gpu");
Int64[] strides = op.GetAttrIntList("strides");
bool useCudnn = op.GetAttrBool("use_cudnn_on_gpu");
String padding = op.GetAttrString("padding");
}
foreach (Output output in op.Outputs)
{
int[] shape = resnet.Graph.GetTensorShape(output);
if (output.NumConsumers == 0)
{
couldBeOutputs.Add(name);
}
}
Buffer buffer = resnet.Graph.GetOpDef(op.OpType);
Tensorflow.OpDef opDef = Tensorflow.OpDef.Parser.ParseFrom(buffer.Data);
}
using (Buffer versionDef = resnet.Graph.Versions())
{
int l = versionDef.Length;
}
Resnet.RecognitionResult[][] results = resnet.Recognize(imageTensor);
}
}
public TensorShapeProto as_proto()
{
TensorShapeProto dim = new TensorShapeProto();
return(new TensorShapeProto(dim));
}
public Shape(TensorShapeProto proto)
{
_dims = proto.Dim.Select(x => x.Size).ToArray();
}
