Work/2026
Kamiru Animation Pipeline
Computer vision that digitizes a real artist’s analog animation — scans straighten, identify and crop themselves.
256-bitdHash frame dedup
16-bitend-to-end color pipeline
losslessFFmpeg extraction & rebuild
A cross-platform desktop app that digitizes a real artist’s mixed-media and cyanotype animation workflow: a video becomes printable contact sheets, the artist paints or exposes each frame on paper, and the app takes the scans back — straightens, identifies and crops every frame on its own, then rebuilds the final video, losslessly and in order.
Highlights
- Self-correcting registration. Redundant ArUco markers (paint over half of them and it still works) plus one QR per frame, aligned with RANSAC homography in a coarse-to-fine pass, with the scan’s true scale measured from the markers themselves — any scanner resolution just works.
- Perceptual deduplication. Held frames are detected with a 256-bit dHash and Hamming distance so repeated drawings print once and get reused at rebuild time, saving paper, ink and hours of painting.
- Color-faithful processing. 16-bit end-to-end, lossless PNG extraction via FFmpeg, and calibratable cyanotype negative curves (Easy Digital Negatives method) with ink-saving modes.
