[C#] ML.NET: ArgumentOutOfRangeException thrown in PredictionEngine.Predict

[abhaytechnoscore]

Describe the issue

public class HomeController : Controller
{
private readonly MLContext _mlContext;
private readonly PredictionEngine<SignLanguageInput, SignLanguageOutput> _predictionEngine;

public HomeController()
{
    _mlContext = new MLContext();

    // Load ONNX model 
    var modelPath = Path.Combine(Directory.GetCurrentDirectory(), "wwwroot", "models", "hand_landmark_sparse_Nx3x224x224.onnx"); 

    var dataView = _mlContext.Data.LoadFromEnumerable(new List<SignLanguageInput>());
    var pipeline = _mlContext.Transforms.ApplyOnnxModel(modelPath);
      
    var trainedModel = pipeline.Fit(dataView);
    _predictionEngine = _mlContext.Model.CreatePredictionEngine<SignLanguageInput, SignLanguageOutput>(trainedModel);
}

public IActionResult Index()
{
    return View();
}
 
public IActionResult PredictGesture([FromBody] string base64Image)
{
    if (string.IsNullOrEmpty(base64Image) || base64Image == "data:,")
        return BadRequest("Image data is missing!");

    // Decode the base64 image and save it temporarily
    var imageBytes = Convert.FromBase64String(base64Image.Replace("data:image/png;base64,", ""));
    var tempImagePath = Path.Combine(Path.GetTempPath(), $"{Guid.NewGuid()}.png");
    System.IO.File.WriteAllBytes(tempImagePath, imageBytes);

    try
    {
        // Preprocess the image to create input tensor
        var inputTensor = PreprocessImage(tempImagePath, 92, 92);  
        float[] inputArray = inputTensor.ToArray();
        // Create input for prediction
        var input = new SignLanguageInput { input = inputArray };

        // Predict gesture
        var result = _predictionEngine.Predict(input);

        if (result != null)
        {
            return Ok(new
            {
                Prediction = result.Label,
                Confidence = result.Confidence,
            });
        }
        else
        {
            return StatusCode(500, "Prediction returned null");
        }
    }
    catch (Exception ex)
    {
        return StatusCode(500, $"Internal server error: {ex.Message}");
    }
    finally
    {
        // Clean up temporary file
        System.IO.File.Delete(tempImagePath);
    }
}


private DenseTensor<float> PreprocessImage(string imagePath, int width, int height)
{
    using var bitmap = new Bitmap(imagePath);
    using var resized = new Bitmap(bitmap, new Size(width, height));  

    int channels = 3; // RGB
    var tensor = new float[channels * width * height];  
    int index = 0;

    for (int y = 0; y < resized.Height; y++)
    {
        for (int x = 0; x < resized.Width; x++)
        {
            var pixel = resized.GetPixel(x, y);

            // Channel-first format (C, H, W)
            tensor[index + 0] = pixel.R / 255f; // Red
            tensor[index + 1] = pixel.G / 255f; // Green
            tensor[index + 2] = pixel.B / 255f; // Blue

            index += channels;
        }
    }

    // Create a DenseTensor directly from the preprocessed image
    var tensorShape = new[] { 1, 3, height, width }; // NCHW format
    return new DenseTensor<float>(tensor, tensorShape); // Return as DenseTensor
}

}

I have given my code (in mvc). In this code, i am getting error on line "var result = _predictionEngine.Predict(input);" and error is "System.ArgumentOutOfRangeException: 'Index was out of range. Must be non-negative and less than the size of the collection. (Parameter 'index')'"

Using package :

.Net Framework : net8.0
Microsoft.ML : Version="4.0.0"
Microsoft.ML.OnnxRuntime.Gpu Version="1.20.1"
Microsoft.ML.OnnxRuntime.Managed Version="1.20.1"
Microsoft.ML.OnnxTransformer Version="4.0.0"
SixLabors.ImageSharp Version="3.1.6"
System.Drawing.Common Version="9.0.0"

and using window is "Window 11" with 64bit OS

To reproduce

to reproduce it, plz run application and give sign image and then it will give error.

Urgency

No response

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

1.20.1

Execution Provider

Other / Unknown

[Jeffinp]

The Problem: System.ArgumentOutOfRangeException

This error almost always means your input data doesn't match what your ONNX model expects.

The Solution (Focus on these):

1. Image Size (99% likely the issue):

   • Your model hand_landmark_sparse_Nx3x224x224.onnx is designed for 224x224 pixel images.
   • You're resizing to 92x92 in PreprocessImage. This is incorrect.
   • Fix: Change your PreprocessImage function to resize to 224x224:

   private DenseTensor<float> PreprocessImage(string imagePath, int width, int height)
   {
       using var bitmap = new Bitmap(imagePath);
       using var resized = new Bitmap(bitmap, new Size(224, 224)); // Change to 224x224
   
       // ... (rest of your code)
   
       var tensorShape = new[] { 1, 3, 224, 224 }; // Change to 224x224
       return new DenseTensor<float>(tensor, tensorShape);
   }

   Also update the PredictGesture:

   public IActionResult PredictGesture([FromBody] string base64Image)
   {
       // ... (rest of your code)
   
       try
       {
           // Preprocess the image to create input tensor
           var inputTensor = PreprocessImage(tempImagePath, 224, 224); // Update image size
           // ... (rest of your code)
       }
       // ... (rest of your code)
   }

2. Output Column Names:

   • Your ONNX model might use different output column names than just Label and Confidence.
   • Fix: Use Netron (a tool for visualizing neural network models) to open your .onnx file. Check the exact output names.
   • Update these names in your ApplyOnnxModel call:

   var pipeline = _mlContext.Transforms.ApplyOnnxModel(
        outputColumnNames: new[] { "correct_label_name", "correct_confidence_name" }, // Replace with the actual names
        inputColumnNames: new[] { "input" },
        modelFile: modelPath);

3. SignLanguageInput and SignLanguageOutput:

   • Double-check that the Label and Confidence properties in these classes match the exact output names from your model.
   • Make sure the VectorType attribute correctly reflects the dimensions: [1, 3, 224, 224].

   public class SignLanguageInput
   {
       [ColumnName("input")] // Should be "input"
       [VectorType(1, 3, 224, 224)]
       public float[] input { get; set; }
   }
   
   public class SignLanguageOutput
   {
       [ColumnName("correct_label_name")] // Replace with the actual name
       public string Label { get; set; }
   
       [ColumnName("correct_confidence_name")] // Replace with the actual name
       public float[] Confidence { get; set; }
   }

In short, do this:

1. Resize your image to 224x224 in PreprocessImage.
2. Use Netron to find the correct output names from your model and update your code accordingly.
3. Make sure your SignLanguageInput and SignLanguageOutput classes are defined correctly.

After making these changes, the error should be gone. If you still have problems, use Console.WriteLine to print the shape and data type of your tensor to help debug further.

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.