Video feature tracking with SIFT and calcOpticalFlowPyrLK?

I’m trying to track features on a live video(at 60fps or higher) and ideally use SIFT for feature detection.

Is periodically detecting keypoints with SIFT (say every 15 frames) and then calculating LC-Optical flow on all intermediate frames the best wait to track keypoints on an intricate object?

I was initially trying to use SIFT at 100fps but that hasn’t worked and tracking alone loses points after like 200 frames using this demo: opencv/ at 4.x · opencv/opencv · GitHub

I’d appreciate any other ideas on tracking at a high frame rate.