fullseye

fly_hs_readout — FLYVISION integrate op

使い方

Horizontal-system wide-field readout: opponent sum over the upper field.

Given a stack of motion-detector responses over the ommatidia — the first n_pref rows the preferred-direction type, the rest the anti-preferred — the opponent readout over the ommatidia above the horizon is::

(a - b) / (a + b + 1e-6)

where a and b are the (optionally rectified) totals of the preferred and anti-preferred responses over the columns with el > el_min_deg.

Only the upper visual field is used because the ground flow of forward motion is 1/distance while the rotational flow it must be separated from is distance-independent: an eye 1.2 mm above the floor sees floor flow of 330-1000 deg/s, above the ~200 deg/s bandwidth of the motion detectors, so the lower field carries speed the correlator cannot read and is excluded.

responses: a (k, n) matrix of responses, n matching the lattice. lattice: the dict returned by :func:fly_hex_lattice. n_pref: how many leading rows are the preferred-direction type (1 <= n_pref < k). el_min_deg: the horizon; only ommatidia above it contribute. rectify: half-wave rectify (clip negatives to 0) before summing.

Returns the opponent ratio as a float in [-1, 1].

Ground truth: with the preferred rows active only in the upper field and the anti-preferred rows silent, the readout is +1 (and the sign and value match the hand calculation); an el_min_deg above every ommatidium raises (pinned in the tests).

Raises ValueError: a non-2-D / empty / non-finite responses, a responses over the element cap, a lattice that is not a :func:fly_hex_lattice result, a column count not matching the lattice, a n_pref outside [1, k-1], a non-real el_min_deg, a non-bool rectify, and an el_min_deg above every ommatidium (no upper field to read).

参考(サンプルデータ・文献)

実行できる例(この op を実際に呼ぶ検証済みサンプル)

型が繋がる次の op(measurement を入力に取れる)

同カテゴリ(integrate)


Provenance: flyvision.py — FLYVISION operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.