flow op• Data kinds: voxel × voxel → flow_dense
• Call: import fullseye as fs; fs.ledger.vol_flow_3d(vol_a, vol_b, win: 'int' = 3, pyr_levels: 'int' = 3, iters: 'int' = 5, reg: 'float' = 0.001) -> 'np.ndarray' (to call the implementation directly, import live4d; live4d.vol_flow_3d(vol_a, vol_b, win: 'int' = 3, pyr_levels: 'int' = 3, iters: 'int' = 5, reg: 'float' = 0.001) -> 'np.ndarray'; from the registry, opslive4d.get("vol_flow_3d"))
> This operator's description has not been translated yet. The original text follows as it is.
2 つの体積の間の密な変位場 `(3, Z, Y, X)(flow_dense`、成分 dz, dy, dx [voxel])。
`vol_b(x) ≈ vol_a(x − d)、つまり vol_a の構造が +d 動いて vol_b` になる(torch 版
`scene_flow_lk と同じ約束)。Lucas–Kanade の 3 次元版を numpy + scipy だけで: win` は窓の半幅
(窓は一辺 2·win + 1)、`pyr_levels はピラミッド段数(大変位ほど増やす、各段 1/2)、iters` は各段の
warp 反復、`reg` は構造テンソル対角の正則化(平坦部の 0 除算回避。大きいほど平坦部の流れが 0 に寄る)。
テクスチャの無い平坦部では流れは決まらない(開口問題)—— 信じるのは勾配のある場所だけ。
>>> a = volseq_synth_dividing((16, 24, 24), n_frames=4, split_frame=0, speed=1.0)
>>> d = vol_flow_3d(a[1], a[2]); d.shape
(3, 16, 24, 24)
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_live4d — py -3.11 examples/poc_live4d.py
flow_dense as input)—
flow)volseq_speed · volseq_pathline_render · volseq_pathline_orbit
*Provenance: live4d.py — LIVE4D operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.