Ground truth vs. estimator predictions, rendered live on the raw frames (no pre-baked overlays).
Navigate with ← / → or the Prev/Next buttons (the list loops). z toggles the GT keypoints, x the prediction keypoints. 1–9, 0 (=10th) select the predictor at that position in the dropdown list. Canvases render at a fixed 640×480 internal resolution (matching the source frames); CSS scales them for display, so annotations stay pixel-aligned at any window size.