OpenCV PCACompute_3

PCACompute_3 is the UMat twin of PCACompute_2: same signature, same cv2.PCACompute call, no behavioral difference. The full explanation lives in the PCACompute_2 article - this is the short version for anyone who landed on this number first.

What this node does: it learns a PCA model the "variance" way. You hand it a dataset and a retainedVariance fraction, and it returns the mean and eigenvectors, keeping exactly as many principal components as needed to explain that fraction of the data's variance. 0.95 = "keep enough components that I get 95% of the variance back." OpenCV decides the component count; you just set the quality bar. That's a more intuitive dial than the maxComponents integer used by the _0/_1 twins.

The returned mean + eigenvectors are the model that feeds PCAProject_0 (compress) and PCABackProject_0 (reconstruct) - the encode/decode halves of the PCA pipeline.

The inputs that matter

• data (NPARRAY) - training samples, one per row.
• mean (NPARRAY) - required by this generated wrapper even though raw OpenCV can compute it from an empty array. Zero vector works as a "no centering" stand-in.
• retainedVariance (FLOAT) - fraction of variance to preserve, 0–1. 0.9–0.99 is the sane band.
• eigenvectors (NPARRAY, optional) - OpenCV out-parameter; ignore it.

Outputs: nparray_0 (mean), nparray_1 (eigenvectors).

How to install it

Pack-level install, once:

• ComfyUI Manager → search opencv-comfyui → Install, restart ComfyUI.
• Or manually:

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

Requires opencv-contrib-python. No model downloads.

Common issues & troubleshooting

• retainedVariance out of (0, 1] is the classic mis-set. Keep it in range.
• Shape errors: data and mean column counts must match ((-215:Assertion failed)).
• Batch rule: batch_size==1 only - pull a frame with ImageFromBatch if needed.

Pick _2 or _3; they're the same node. And if you'd rather specify component count than variance, the _0/_1 twins do exactly that.