FeatureLab Mini — Classic vs DL

Minimal feature-detection demo with Classical and Deep Learning paths, now backed by a shared runtime.
Detector
Line detector (classical)

Choose LSD to enable OpenCV's Line Segment Detector instead of Probabilistic Hough.
0 255
0 255
0.02 0.15
2 8
3 7
1 200
1 300
0 50
10 50000
1 20

Enable subpixel iso-contours (marching squares) for DexiNed outputs.
Threshold mode

Adaptive uses mean + σ·std; fixed uses a constant probability threshold.
0 5
0 1
-0.5 0.5
                    Place ONNX files in `./models` (create the folder next to the repo root).

                    **Expected filenames (defaults):**
                    - **Edges (Canny)**: hed.onnx, dexined.onnx

  • Corners (Harris): superpoint.onnx

  • Lines (Hough/LSD): sold2.onnx, hawp.onnx

  • Ellipses (Contours + fitEllipse): ellipse_head.onnx

                      Backends use onnxruntime with CoreML (if available) or CPU provider.