fly_flow_from_directions — FLYVISION direction op

• 데이터 종류: matrix × table → matrix

• 호출: import fullseye as fs; fs.ledger.fly_flow_from_directions(responses, lattice)(구현을 직접 호출하려면 import flyvision; flyvision.fly_flow_from_directions(responses, lattice), 원장에서 가져오려면 opsflyvision.get("fly_flow_from_directions"))

사용법

> 이 연산자의 설명은 아직 번역이 없습니다. 원문을 그대로 싣습니다.

Six directional responses per ommatidium in, one local flow vector out.

The lobula plate does not keep six numbers per point of the visual field; it

keeps the direction and strength of the motion there, split over four layers

of opposite preference. This op is that reduction, done as the vector sum on

the tangent plane that makes opposite hexagonal directions cancel exactly::

f_i = (2/6) * sum_k R[k, i] * (cos t_ik, sin t_ik)

where `t_ik` is the angle of the step to ommatidium *i*'s neighbour in

direction *k*, measured in the local tangent plane with azimuth scaled by

`cos(elevation)` so that it is an angle on the sphere rather than a

difference of coordinates. The `2/6` normalises the sum so that a

cosine-tuned set of responses of amplitude `A` returns a vector of length

`A — with six directions the raw sum is 3A`.

responses: `(6, n), the return of :func:fly_t4t5_field`.

lattice: the :func:fly_hex_lattice it was measured on (a merged eye is

refused — neighbours are local, the fit downstream is not).

Returns `(n, 2)` float64: for each ommatidium the flow component along the

local azimuth (left positive) and along the local elevation (up positive), in

the response's own units. This is the `(n, 2)` field

:func:fly_egomotion_from_flow and :func:fly_matched_filter speak.

Ommatidia at the rim, which do not have all six neighbours, return exactly

zero: a partial sum over the circle does not cancel, so keeping it would draw

a ring of inward flow around the eye and the fit downstream would read that

ring as a rotation.

Ground truth: for responses `R[k] = A cos(theta_k - theta_0)` the returned

vector has length `A and angle theta_0` exactly (regular geometry, where

the six steps are 60 degrees apart); opposite directions cancel, so a

symmetric flicker response of any size returns exactly zero.

Raises `ValueError: *responses* that is not (6, n)`, a column count

that is not the lattice's ommatidium count, non-finite entries, and any

malformed *lattice*.

자세한 사용 가이드

• fly_vision 패밀리 가이드

참고(샘플 데이터·문헌)

• 샘플 데이터 카탈로그(DL URL / 라이선스) —— 2-D 는 skimage.data(BSD/public)+ 합성, 3-D 는 실데이터 소스(Stanford/PDS 등)의 DL URL.

• 연산자의 내력·참고문헌 —— 이 연산자 족의 바탕이 된 연구/기법의 출처.

• 알고리즘의 정전(저자·연도)과 용도는 위의 패밀리 사용 가이드에 적혀 있습니다.

실행 가능한 예제(이 연산자를 실제로 호출하는 검증된 샘플)

• poc_fly_optomotor_steering — py -3.11 examples/poc_fly_optomotor_steering.py

타입이 이어지는 다음 연산자(matrix 를 입력으로 받는 것)

fly_lamina_filter · fly_onoff_split · fly_t4t5_field · fly_egomotion_from_flow · fly_hs_readout

같은 카테고리(direction)

fly_t4t5_field


*Provenance: flyvision.py — FLYVISION 연산자 레지스트리. 이 op 노트는 tools/opdocs.py md 가 자동 생성합니다(직접 편집하지 마세요).*

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