OpenCV reprojectImageTo3D_1

reprojectImageTo3D_1 is the duplicate of reprojectImageTo3D_0 - the same cv2.reprojectImageTo3D overload emitted twice by opencv-comfyui's (geroldmeisinger/opencv-comfyui) generator. Identical inputs, identical output, no hidden difference. Bookmark the _0, ignore the _1.

For the record, both do the same thing: convert a stereo disparity map into per-pixel (X, Y, Z) world coordinates. The inputs are disparity (a single-channel NPARRAY from a stereo matcher), Q (a 4×4 calibration matrix as an NPARRAY), handleMissingValues (boolean), ddepth (int, usually 5 for CV_32F), and an optional _3dImage out-parameter you can leave alone. The output is a three-channel array of float coordinates - data, not a viewable image, so it doesn't belong in Nparrays2Image.

The longer story - why this is the payoff node of real stereo vision, why it needs a calibrated rig rather than a text prompt, and why the neural single-image depth packs are the better tool for most people - is in the reprojectImageTo3D_0 article. Read that one; it's the whole essay.

Installing it

The standard pack install. ComfyUI Manager (search "opencv-comfyui"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui

restart, confirm OpenCV:

pip install opencv-contrib-python

And the recurring pack rule one more time: the trailing number is an overload counter from the code generator, not a version. When _0 and _1 look the same, they are the same.