[rsasaki0109]

The README's two image animations previously ran Rust on synthetic checkerboard frames, while the depth-sweep card only illustrated geometry. Replace those three cards with actual Rust ORB detections and dense integer flow on a pinned real TUM camera recording, and actual RANSAC plane removal on the public PCL scan.

Decode the first 64 consecutive original movie frames and compute all 63 adjacent flow pairs with SpatialRust. Render every second result and independent ORB detections; OpenCV only decodes the movie. Draw returned coordinates/vectors at the image scale, omit non-finite vectors, and preserve per-frame input/output hashes plus CC BY 4.0 attribution. Use a shared photographic palette; keep all nine GIFs under the existing 3 MB cap (2,866,446 bytes).

Validation: independently recomputed every displayed pair and matched all 32 ORB coordinate/descriptor and dense-flow hashes; decoded all 304 GIF frames; checked hashes/looping/size; all nine images loaded without overflow at 1100, 390 and 320 px; inspected real-video previews; git diff --check passed. No production API, native dependency or estimator changes. Only the 8 MB official RGB movie was acquired; no RGB-D archive, build or model download.

ORB detections are per-frame features, and flow vectors are raw block matches; the GIFs make no object-ID, calibrated-time, GPU speed, metric pose or SLAM accuracy claim. Source, license and reproducible commands are documented in docs/README_VISUALS.md.

The camera-rotation browser control now waits for both target intrinsic dimensions before releasing a held bitmap, and snapshots its frame cursor together with the resize. All existing stale-geometry assertions remain. The complete 106-control browser suite passed locally with zero console errors and zero external responses against the hash-verified published app build; production camera code is unchanged.