OpenCV convexityDefects_1

convexityDefects_1 is the second overload of cv2.convexityDefects - the UMat variant next to the MatLike one in OpenCV's type definitions - and this auto-generated pack shipped both. The nodes are identical. The real article is convexityDefects_0; this is the signpost page for the _1 suffix, plus the one caveat worth repeating.

What the node does: given a contour (NPARRAY) and a convexhull (NPARRAY), it finds the indentation points where the contour diverges from its convex hull - the valleys between fingers in hand tracking, the notches in an irregular detected region. Output is an NPARRAY of (N, 1, 4) rows: [start_index, end_index, farthest_index, depth], where depth is how far the contour sags from the hull and the one column you'll actually threshold on.

The caveat, because it's the classic failure: the convexhull input must be the index form, produced by convexHull_0 with returnPoints set to False. Feed it a points-form hull and OpenCV rejects it with an assertion error and zero explanation in the UI. That single detail is the difference between a working finger-counting chain and an afternoon of head-scratching.

Install, same as the pack

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui
pip install opencv-python-contrib

Or ComfyUI Manager → opencv-comfyui → restart.

And the standard nparray reminders: these nodes run on numpy arrays, not Comfy IMAGE tensors - Image2Nparray (batch 1 only) on the way in - and the outputs are geometry, not pictures, so they won't display through Nparrays2Image.

Between _0 and _1: no functional difference whatsoever. If your workflow references convexityDefects_1, keep it; if you're building fresh, pick either. The suffix is pure auto-generation noise from overloaded signatures.