OpenCV PCACompute_1

This is the UMat twin of PCACompute_0. Same inputs (data, mean, maxComponents, optional eigenvectors), same outputs (nparray_0 = mean, nparray_1 = eigenvectors), same call to cv2.PCACompute. The pack's generator emits a node for every OpenCV type declaration, and OpenCV declares PCACompute twice - MatLike and UMat - so you get two identical wrappers. There is no reason to prefer one over the other. If you landed here, the article on PCACompute_0 has the full story; this one covers the same ground with the load-bearing details.

What it does

cv2.PCACompute(data, mean, eigenvectors, maxComponents) computes the mean and principal components of a dataset (rows = samples) and returns (mean, eigenvectors). Those two outputs feed straight into PCAProject_0 (to compress) and PCABackProject_0 (to reconstruct), so this node is the model-learning step that the rest of the PCA family consumes. maxComponents sets how many components you keep - the dial between "tiny and lossy" and "bigger and accurate."

The inputs that matter

• data (NPARRAY) - training samples, one per row.
• mean (NPARRAY) - required here even though OpenCV can compute it internally. Zero vector is a fine "skip centering" stand-in; real mean is better.
• maxComponents (INT) - component count.
• eigenvectors (NPARRAY, optional) - out-parameter, leave it alone.

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 downloads.

Common issues & troubleshooting

• When in doubt, use PCACompute2 (_0/_2 variants) - it also returns eigenvalues and copes better when samples are fewer than dimensions.
• Assertion failures are shapes. data and mean column counts must match.
• Batch rule: batch_size==1 only; use ImageFromBatch if needed.

Same node, different number. The only real decision in the PCACompute family is _0/_1 (max components) vs _2/_3 (retained variance) - and within each pair, the twins are interchangeable.