diff --git a/src/AIFotoONLUS.Core/AIFotoONLUS.Core.xml b/src/AIFotoONLUS.Core/AIFotoONLUS.Core.xml
index 39fc26a..4b6c8f2 100644
--- a/src/AIFotoONLUS.Core/AIFotoONLUS.Core.xml
+++ b/src/AIFotoONLUS.Core/AIFotoONLUS.Core.xml
@@ -139,6 +139,12 @@
must match the class ordering used by the trained recognition network.
+
+
+ When enabled, request OpenCV DNN CUDA backend/target for inference.
+ The installed OpenCV runtime must have CUDA support or model loading/forwarding may fail.
+
+
When enabled, recognition crops will be saved to disk under
diff --git a/src/AIFotoONLUS.Core/ModelConfiguration.cs b/src/AIFotoONLUS.Core/ModelConfiguration.cs
index 0b0b553..51d7ac7 100644
--- a/src/AIFotoONLUS.Core/ModelConfiguration.cs
+++ b/src/AIFotoONLUS.Core/ModelConfiguration.cs
@@ -55,6 +55,12 @@ namespace AIFotoONLUS.Core
///
public string[] NumberClasses { get; set; } = new[] { "0", "1", "2", "3", "4", "5", "6", "7", "8", "9" };
+ ///
+ /// When enabled, request OpenCV DNN CUDA backend/target for inference.
+ /// The installed OpenCV runtime must have CUDA support or model loading/forwarding may fail.
+ ///
+ public bool UseGpu { get; set; } = false;
+
///
/// When enabled, recognition crops will be saved to disk under
/// "logs/crops" for diagnostic inspection. Disabled by default.
diff --git a/src/AIFotoONLUS.Core/NumberRecognitionEngine.cs b/src/AIFotoONLUS.Core/NumberRecognitionEngine.cs
index b9b9284..d6b2d25 100644
--- a/src/AIFotoONLUS.Core/NumberRecognitionEngine.cs
+++ b/src/AIFotoONLUS.Core/NumberRecognitionEngine.cs
@@ -95,10 +95,8 @@ namespace AIFotoONLUS.Core
_detectionNet = CvDnn.ReadNetFromDarknet(_cfg.DetectionCfg, _cfg.DetectionWeights);
_recognitionNet = CvDnn.ReadNetFromDarknet(_cfg.RecognitionCfg, _cfg.RecognitionWeights);
- _detectionNet.SetPreferableBackend(Backend.OPENCV);
- _detectionNet.SetPreferableTarget(Target.CPU);
- _recognitionNet.SetPreferableBackend(Backend.OPENCV);
- _recognitionNet.SetPreferableTarget(Target.CPU);
+ ConfigureNetRuntime(_detectionNet, _cfg.UseGpu);
+ ConfigureNetRuntime(_recognitionNet, _cfg.UseGpu);
// Let OpenCV use multiple threads internally (use number of logical processors)
try
{
@@ -108,6 +106,11 @@ namespace AIFotoONLUS.Core
{
// Ignore if not supported by OpenCvSharp build
}
+
+ if (_cfg.UseGpu)
+ {
+ ValidateGpuRuntime();
+ }
}
public void Dispose()
@@ -119,6 +122,38 @@ namespace AIFotoONLUS.Core
GC.SuppressFinalize(this);
}
+ public static bool TryValidateGpuRuntime(ModelConfiguration cfg, ILogger? logger, out string? failureMessage)
+ {
+ if (cfg is null) throw new ArgumentNullException(nameof(cfg));
+
+ var probeConfiguration = new ModelConfiguration
+ {
+ DetectionCfg = cfg.DetectionCfg,
+ DetectionWeights = cfg.DetectionWeights,
+ RecognitionCfg = cfg.RecognitionCfg,
+ RecognitionWeights = cfg.RecognitionWeights,
+ ConfidenceThreshold = cfg.ConfidenceThreshold,
+ NmsThreshold = cfg.NmsThreshold,
+ DetectionInputSize = cfg.DetectionInputSize,
+ RecognitionInputSize = cfg.RecognitionInputSize,
+ NumberClasses = cfg.NumberClasses,
+ EnableCropSaving = cfg.EnableCropSaving,
+ UseGpu = true
+ };
+
+ try
+ {
+ using var engine = new NumberRecognitionEngine(probeConfiguration, logger);
+ failureMessage = null;
+ return true;
+ }
+ catch (Exception ex)
+ {
+ failureMessage = ex.GetBaseException().Message;
+ return false;
+ }
+ }
+
private static string SanitizeFileName(string name)
{
foreach (var c in Path.GetInvalidFileNameChars()) name = name.Replace(c, '_');
@@ -127,6 +162,39 @@ namespace AIFotoONLUS.Core
private string[] GetOutputLayerNames(Net net) => net.GetUnconnectedOutLayersNames();
+ private static void ConfigureNetRuntime(Net net, bool useGpu)
+ {
+ if (useGpu)
+ {
+ net.SetPreferableBackend(Backend.CUDA);
+ net.SetPreferableTarget(Target.CUDA);
+ return;
+ }
+
+ net.SetPreferableBackend(Backend.OPENCV);
+ net.SetPreferableTarget(Target.CPU);
+ }
+
+ private void ValidateGpuRuntime()
+ {
+ try
+ {
+ using var detectionProbe = new Mat(_cfg.DetectionInputSize.Height, _cfg.DetectionInputSize.Width, MatType.CV_8UC3, Scalar.All(0));
+ _ = DetectTextRegions(_detectionNet, detectionProbe).Take(1).ToArray();
+
+ using var recognitionProbe = new Mat(_cfg.RecognitionInputSize.Height, _cfg.RecognitionInputSize.Width, MatType.CV_8UC3, Scalar.All(0));
+ using var blob = CvDnn.BlobFromImage(recognitionProbe, 0.00392, _cfg.RecognitionInputSize, new Scalar(0, 0, 0), true, false);
+ _recognitionNet.SetInput(blob);
+ using var output = _recognitionNet.Forward();
+ }
+ catch (Exception ex)
+ {
+ throw new InvalidOperationException(
+ "OpenCV DNN CUDA runtime validation failed. Disable number AI GPU mode or use an OpenCV runtime built with CUDA DNN support.",
+ ex);
+ }
+ }
+
///
/// Detect text regions in the supplied image using the detection network.
///
@@ -152,7 +220,7 @@ namespace AIFotoONLUS.Core
var outNames = GetOutputLayerNames(detectionNet);
var outsList = new List();
detectionNet.Forward(outsList, outNames);
-
+
Mat[] outs = outsList.ToArray();
if (outs.Length == 0)
{
@@ -162,15 +230,15 @@ namespace AIFotoONLUS.Core
var fallback = new List();
for (int on = 0; on < outNames.Length; on++)
{
- try
- {
- var single = detectionNet.Forward(outNames[on]);
- fallback.Add(single);
- }
- catch (Exception ex)
- {
- _logger?.LogError(ex, "Fallback Forward failed for {name}", outNames[on]);
- }
+ try
+ {
+ var single = detectionNet.Forward(outNames[on]);
+ fallback.Add(single);
+ }
+ catch (Exception ex)
+ {
+ _logger?.LogError(ex, "Fallback Forward failed for {name}", outNames[on]);
+ }
}
if (fallback.Count > 0)
{
@@ -221,21 +289,21 @@ namespace AIFotoONLUS.Core
}
if (maxScore > _cfg.ConfidenceThreshold)
- {
- int x = (int)Math.Max(0, Math.Round(cx - w / 2));
- int y = (int)Math.Max(0, Math.Round(cy - h / 2));
- var rect = new Rect(x, y, (int)Math.Round(w), (int)Math.Round(h));
- boxes.Add(rect);
- confidences.Add(maxScore);
- classIds.Add(bestClass);
- centerXList.Add(cx);
- }
+ {
+ int x = (int)Math.Max(0, Math.Round(cx - w / 2));
+ int y = (int)Math.Max(0, Math.Round(cy - h / 2));
+ var rect = new Rect(x, y, (int)Math.Round(w), (int)Math.Round(h));
+ boxes.Add(rect);
+ confidences.Add(maxScore);
+ classIds.Add(bestClass);
+ centerXList.Add(cx);
+ }
}
}
if (boxes.Count == 0) return Enumerable.Empty();
-
+
CvDnn.NMSBoxes(boxes, confidences, (float)_cfg.ConfidenceThreshold, (float)_cfg.NmsThreshold, out int[] indices);
@@ -486,10 +554,8 @@ namespace AIFotoONLUS.Core
{
var det = CvDnn.ReadNetFromDarknet(_cfg.DetectionCfg, _cfg.DetectionWeights);
var rec = CvDnn.ReadNetFromDarknet(_cfg.RecognitionCfg, _cfg.RecognitionWeights);
- det.SetPreferableBackend(Backend.OPENCV);
- det.SetPreferableTarget(Target.CPU);
- rec.SetPreferableBackend(Backend.OPENCV);
- rec.SetPreferableTarget(Target.CPU);
+ ConfigureNetRuntime(det, _cfg.UseGpu);
+ ConfigureNetRuntime(rec, _cfg.UseGpu);
netsBag.Add((det, rec));
return (det, rec);
});
@@ -525,8 +591,7 @@ namespace AIFotoONLUS.Core
try
{
using var tempRec = CvDnn.ReadNetFromDarknet(_cfg.RecognitionCfg, _cfg.RecognitionWeights);
- tempRec.SetPreferableBackend(Backend.OPENCV);
- tempRec.SetPreferableTarget(Target.CPU);
+ ConfigureNetRuntime(tempRec, _cfg.UseGpu);
var alt = RecognizeDigits(crop, tempRec, ctx);
if (!string.IsNullOrEmpty(alt)) txt = alt;
}