Enum ObjectDetectorModel

Namespace

VisioForge.Core.Types.X.AI

Assembly

VisioForge.Core.dll

The object-detection model family the detector decodes. Each family has a distinct ONNX output layout and a distinct frame-preprocessing convention (resize mode, pixel normalization, channel order), so the detector must be told which one the supplied .onnx file belongs to.

public enum ObjectDetectorModel

Fields

YOLOv8 = 0

Ultralytics YOLOv8 / YOLO11 layout: a single transposed tensor [1, 4 + numClasses, numAnchors] (box values followed by per-class scores, no objectness term). Decoded with class-wise NMS. Frames are letterboxed (centered, gray 114 padding), pixels normalized to 0..1, channel order RGB.

YOLOX = 1

YOLOX (Megvii, Apache-2.0) layout: a single tensor [1, numAnchors, 5 + numClasses] (box values, objectness, per-class scores). Decoded with class-wise NMS. Frames are letterboxed (top-left, gray 114 padding), pixels NOT normalized, channel order BGR.

RTDETR = 2

RT-DETR / D-FINE (Apache-2.0) transformer layout as exported by Hugging Face Transformers / onnx-community: a single image input and two float outputs — class logits[1, numQueries, numClasses] (apply sigmoid) and pred_boxes[1, numQueries, 4] (cx, cy, w, h normalized to 0..1). End-to-end / NMS-free. Frames are resized directly to the model input size, pixels normalized to 0..1, channel order RGB.

Remarks

Licensing note: VisioForge.Core.Types.X.AI.ObjectDetectorModel.YOLOv8 decodes Ultralytics YOLOv8/v11 models, which are published under AGPL-3.0 — using them in a closed-source product requires a commercial Ultralytics license, which is the integrator's responsibility. VisioForge.Core.Types.X.AI.ObjectDetectorModel.YOLOX and VisioForge.Core.Types.X.AI.ObjectDetectorModel.RTDETR decode Apache-2.0 model families.