ONNX model inference issues with ML.Net

[crazyoutlook]

System Information (please complete the following information):

• OS & Version: Windows 10
• ML.NET Version: ML.NET v1.6, ML.OnnxRuntime v1.7
• .NET Version: NET 6.0

Describe the bug
I have an ONNX model which is exported from Azure custom vision. It has single output. My ML.Net code pipeline works fine with this ONNX model with single output. I have another Yolo v4 model which I converted to ONNX but post conversion it has dual output. Due to this, my ML.Net code pipeline is not working. I checked the output details using Netron tool.

I used this link to convert Yolo to ONNX: https://github.com/Tianxiaomo/pytorch-YOLOv4

To Reproduce
Steps to reproduce the behavior:

Use ML.Net to infer a ONNX model

Expected behavior
ML.Net should give inference results

Screenshots, Code, Sample Projects

var pipeline = mlContext.Transforms.ResizeImages(resizing: ImageResizingEstimator.ResizingKind.Fill, outputColumnName: "data", imageWidth: ImageSettings.imageWidth, imageHeight: ImageSettings.imageHeight, inputColumnName: nameof(ImageInput.Image))
.Append(mlContext.Transforms.ExtractPixels(outputColumnName: "data", orderOfExtraction: ImagePixelExtractingEstimator.ColorsOrder.AGRB))
.Append(mlContext.Transforms.ApplyOnnxModel(modelFile: modeltoInfer.ModelPath, outputColumnName: "model_outputs0", inputColumnName: "data", gpuDeviceId: gpuDeviceId, fallbackToCpu: fallbackToCpu));

Screenshots:

Additional context
Is there any ML.Net code which can do inferencing for a ONNX model having dual output ?

[luisquintanilla]

Hi @crazyoutlook

Thanks for your question. Here is a sample using an ONNX version of the v2 YOLO model. Hope this helps.