OpenCV decomposeProjectionMatrix_0

OpenCV decomposeProjectionMatrix_0 takes a 3×4 camera projection matrix and decomposes it into everything a computer-vision person could want: the intrinsic camera matrix, the rotation matrix, the translation vector, and the Euler angles. In plain terms: given a single matrix that describes "how a camera sees the world," this node hands you where the camera is, which way it's pointing, and what its lens does to the light.

This is calibration/reconstruction math, not image processing - and it's about as far from "generative" as this pack gets. It exists because geroldmeisinger/opencv-comfyui auto-generates a node for every top-level cv2 function, and cv2.decomposeProjectionMatrix is one of ~635. If you're building augmented-reality, pose-estimation, or 3D-reconstruction pipelines on top of ComfyUI (matching the KB's broader deterministic-post-processing philosophy: no model, just math), this is the node.

How it works

A projection matrix P (3×4) maps 3D world points to 2D image points, and it's the product of the intrinsic matrix and the extrinsic [rotation | translation] pair. decomposeProjectionMatrix factors P into:

• cameraMatrix - the 3×3 intrinsics (focal length, principal point).
• rotMatrix - the 3×3 rotation from world to camera coordinates.
• transVect - the 3×1 translation.
• rotMatrixX, rotMatrixY, rotMatrixZ - 3×3 rotation sub-matrices per axis.
• eulerAngles - 3×1 Euler angles derived from the rotation, useful when you want to know "the camera is tilted 12° down" instead of reading a matrix.

That's the whole output list, and it's why this node has seven outputs - one per return value.

Inputs and outputs

• projMatrix (NPARRAY) - the 3×4 projection matrix. Float matrix, not an image - don't wire Image2Nparray into this and expect anything sensible.
• cameraMatrix, rotMatrix, transVect, rotMatrixX/Y/Z, eulerAngles (NPARRAY, optional) - out-parameters; leave them unconnected, the function fills them and you read the outputs.
• nparray_0 through nparray_6 (outputs) - intrinsics, rotation, translation, axis rotations, Euler angles, in that order.

decomposeProjectionMatrix_1 is the identical MatLike/UMat twin.

Install

ComfyUI Manager (search opencv-comfyui) or:

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

Restart ComfyUI. No models. Use opencv-contrib-python (the README's opencv-python-contrib is the wrong package name); conflicting OpenCV installs show up as Cannot import name 'guidedFilter' from 'cv2.ximgproc', with a known fix.

The trap

Seven float matrices come out, and ComfyUI will happily try to preview any of them as a picture - that's how you get the 'NoneType' object has no attribute 'shape' error from Nparrays2Image, or a screen of static. These outputs are data for other camera nodes, not images. And like everything in this pack, inputs must be batch_size==1 - for matrices that's trivially satisfied, since you're building them yourself.