This folder includes Detic for open‑vocabulary detection/segmentation. The scene is Foundation/FoundationSample.scene.unity, controller Foundation/AiliaFoundationSample.cs.
- Detic (Swin‑B 896, LVIS/COCO)
- Purpose: Open‑vocabulary detection and segmentation with very large label sets (up to ~21K categories).
- Source:
./AiliaFoundationSample.cs - Script behavior: Loads ONNX, runs inference on resized frames, and draws boxes/masks with labels from
AiliaFoundationLabel.cs.
- Downloads
Detic_C2_SwinB_896_4x_IN-21K+COCO_lvis_op16.onnx(+.prototxt) toApplication.temporaryCachePath/detic.
- Open model and set inputs: image tensor (CHW float32) and
[h,w]tensor; callUpdate(). - Read
boxes/scores/classesand per‑categorymask; scale boxes and overlay masks; label withAiliaFoundationLabel.
- Controller:
./AiliaFoundationSample.cs - Labels:
./AiliaFoundationLabel.cs