public ResolutionModelDetails GetResolutionModel(int modelId)
{
try
{
MachineLearningModel machineLearningModel = machineLearningModelDAO.Find(modelId);
IList <byte> upscaleFactors = GetUpscaleFactors(modelId);
if (upscaleFactors.Any())
{
ResolutionModelDetails resolutionDetails = new()
{
Id = machineLearningModel.Id,
Category = machineLearningModel.Name,
Title = $"{machineLearningModel.Architecture} {machineLearningModel.Loss}",
Description = machineLearningModel.LongDescription,
UpscaleFactors = upscaleFactors,
RelatedResolutionModels = new List <RelatedResolutionModel>()
};
foreach (MachineLearningModel model in machineLearningModelDAO.FindByName("Super Resolution"))
{
if (model.Id != modelId)
{
resolutionDetails.RelatedResolutionModels.Add(new RelatedResolutionModel()
{
Id = model.Id,
Title = $"{model.Architecture} {model.Loss}",
UpscaleFactors = GetUpscaleFactors(model.Id)
});
}
}
return(resolutionDetails);
}
else
{
throw new NotFoundException(modelId);
}
}
catch (InstanceNotFoundException)
{
throw new NotFoundException(modelId);
}
}
public DatasetMetrics GenerateDatasetMetrics(int modelId, byte upscaleFactor, IList <byte[]> dataset)
{
try
{
if (dataset.Any())
{
DatasetMetrics datasetMetrics = new()
{
ImageMetrics = new List <ImageMetrics>()
};
switch (modelId)
{
case NearestNeighbor:
datasetMetrics = new()
{
Id = NearestNeighbor,
Title = "Nearest Neighbor",
ImageMetrics = new List <ImageMetrics>()
};
break;
case Bilinear:
datasetMetrics = new()
{
Id = Bilinear,
Title = "Bilinear",
ImageMetrics = new List <ImageMetrics>()
};
break;
case Bicubic:
datasetMetrics = new()
{
Id = Bicubic,
Title = "Bicubic",
ImageMetrics = new List <ImageMetrics>()
};
break;
default:
MachineLearningModel machineLearningModel = machineLearningModelDAO.Find(modelId);
datasetMetrics = new()
{
Id = machineLearningModel.Id,
Title = $"{machineLearningModel.Architecture} {machineLearningModel.Loss}",
ImageMetrics = new List <ImageMetrics>()
};
break;
}
foreach (byte[] image in dataset)
{
Bitmap reference = Converter.ConvertByteArrayToBitmap(image);
Bitmap input = ResizeBitmap(reference, reference.Width / upscaleFactor, reference.Height / upscaleFactor, InterpolationMode.Bicubic);
Bitmap compare = Resize(modelId, upscaleFactor, input);
datasetMetrics.ImageMetrics.Add(
new ImageMetrics()
{
Image = Converter.ConvertBitmapToByteArray(compare),
MSE = Math.Round(Metric.MSE(reference, compare), 4),
PSNR = Math.Round(Metric.PSNR(reference, compare), 4),
SSIM = Math.Round(Metric.SSIM(reference, compare), 4)
}
);
}
datasetMetrics.MSE = Math.Round(datasetMetrics.ImageMetrics.Average(x => x.MSE), 4);
datasetMetrics.PSNR = Math.Round(datasetMetrics.ImageMetrics.Average(x => x.PSNR), 4);
datasetMetrics.SSIM = Math.Round(datasetMetrics.ImageMetrics.Average(x => x.SSIM), 4);
return(datasetMetrics);
}
else
{
throw new ArgumentException("Dataset can not be empty");
}
}
catch (InstanceNotFoundException)
{
throw new NotFoundException(modelId, upscaleFactor);
}
catch (OnnxRuntimeException e)
{
throw new InternalErrorException(e);
}
}
