public static void PrintReuslt(ImageNetDataProbability predictionResult)
{
var defaultForeground = Console.ForegroundColor;
var labelColor = ConsoleColor.Magenta;
var probColor = ConsoleColor.Cyan;
var exactLabel = ConsoleColor.Green;
var failLabel = ConsoleColor.Red;
Console.Write("ImagePath: ");
Console.ForegroundColor = labelColor;
Console.Write($"{Path.GetFileName(predictionResult.ImagePath)}");
Console.ForegroundColor = defaultForeground;
Console.Write(" labeled as ");
Console.ForegroundColor = labelColor;
Console.Write(predictionResult.Label);
Console.ForegroundColor = defaultForeground;
Console.Write(" predicted as ");
if (predictionResult.Label.Equals(predictionResult.PredicateLabel))
{
Console.ForegroundColor = exactLabel;
Console.Write($"{predictionResult.PredicateLabel}");
}
else
{
Console.ForegroundColor = failLabel;
Console.Write($"{predictionResult.PredicateLabel}");
}
Console.ForegroundColor = defaultForeground;
Console.Write(" with probability ");
Console.ForegroundColor = probColor;
Console.Write(predictionResult.Probability);
Console.ForegroundColor = defaultForeground;
Console.WriteLine("");
}
static void Main(string[] args)
{
// 创建上下文
var mlContext = new MLContext();
var loader = new TextLoader(mlContext, new TextLoader.Arguments()
{
Column = new[]
{
new TextLoader.Column("ImagePath", DataKind.Text, 0),
},
HasHeader = false
});
// 读取数据
var data = loader.Read(new MultiFileSource(DataPath));
// 构建评估管道，调整图像使其适应神经网络预期的格式
var pipeline =
mlContext.Transforms.LoadImages(imageFolder: ImageFolder, columns: ("ImagePath", "ImageReal"))
.Append(mlContext.Transforms.Resize("ImageReal", "ImageReal", ImageNetSettings.ImageWidth,
ImageNetSettings.ImageHeight))
.Append(mlContext.Transforms.ExtractPixels(new[]
{
new ImagePixelExtractorTransform.ColumnInfo("ImageReal", "input",
interleave: ImageNetSettings.ChannelsLast, offset: ImageNetSettings.Mean),
}))
.Append(mlContext.Transforms.ScoreTensorFlowModel(TFModelPath, new[] { "input" },
new[] { "softmax2" }));//使用TensorFlow Model去评分，需要借助Netron去查询模型的输入输出节点参数
// 管道适配评估
var model = pipeline.Fit(data);
//预测
var predictionFunc = model.MakePredictionFunction <ImageNetData, ImageNetPrediction>(mlContext);
var testData = ImageNetData.ReadFromCsv(DataPath, ImageFolder);
var labels = File.ReadAllLines(TFModelLabelPath);
foreach (var sample in testData)
{
var probs = predictionFunc.Predict(sample).PredictedLabels;
var imageData = new ImageNetDataProbability()
{
ImagePath = sample.ImagePath,
Label = sample.Label
};
(imageData.PredicateLabel, imageData.Probability) = GetBestLabel(labels, probs);
PrintReuslt(imageData);
}
}