CV Feature Extract (Model)

Runs an ONNX feature extractor model (DISK, SuperPoint, etc.) on an image and outputs keypoints + descriptors in the standard format for 'CV Match Features (Model)' and 'CV Draw Matches'. All .onnx files in models/onnx (including subdirectories) are listed — pick the right extractor for your matcher. Output parsing supports 3-output models (e.g. DISK: keypoints (1,N,2) + scores (1,N) + descriptors (1,N,D)), flat 2-output models ((1,N,4) + (1,N,D)), and dense score/descriptor maps ((1,C,H,W) + (1,C,H,W)). Preprocessing follows the LightGlue reference implementation and is read off the MODEL, not guessed: pixels are scaled to 0..1, fed as RGB or gray according to the channel count the model declares, and padded up to a multiple of 16 (DISK's U-Net needs it). Keypoints always come back in the INPUT image's pixels. Zero features is a valid result (empty outputs, not an error).