Skip to content

WIP: reduced precision for lanczos_opencv #635

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 4 commits into
base: master
Choose a base branch
from
Draft

WIP: reduced precision for lanczos_opencv #635

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a127d7d
reduced precision for lanczos_opencv
bjarthur Oct 24, 2025
5f69ead
function barrier for float(F)
bjarthur Oct 24, 2025
6dfc814
specify lanczos precision in type parameter
bjarthur Oct 27, 2025
740fa93
more faithfully follow opencv's lanczos code
bjarthur Oct 27, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 13 additions & 13 deletions src/lanczos/lanczos_opencv.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,11 +50,13 @@ export Lanczos4OpenCV
Alternative implementation of Lanczos resampling using algorithm `lanczos4` function of OpenCV:
https://github.com/opencv/opencv/blob/de15636724967faf62c2d1bce26f4335e4b359e5/modules/imgproc/src/resize.cpp#L917-L946
"""
struct Lanczos4OpenCV <: AbstractLanczos end
struct Lanczos4OpenCV{T} <: AbstractLanczos end
Lanczos4OpenCV() = Lanczos4OpenCV{Float64}()

degree(::Lanczos4OpenCV) = 4

value_weights(::Lanczos4OpenCV, δx) = _lanczos4_opencv(δx)
value_weights(::Lanczos4OpenCV{T}, δx) where {T} =
_lanczos4_opencv(float(T), float(T).(l4_2d_cs), δx)

# s45 = sqrt(2)/2
const s45 = 0.70710678118654752440084436210485
Expand All @@ -65,20 +67,18 @@ const s45 = 0.70710678118654752440084436210485
const l4_2d_cs = SA[1 0; -s45 -s45; 0 1; s45 -s45; -1 0; s45 s45; 0 -1; -s45 s45]


function _lanczos4_opencv(δx)
p_4 = π / 4
y0 = -(δx + 3) * p_4
function _lanczos4_opencv(::Type{F}, l4_2d_cs, δx) where {F}
p_4 = π * F(0.25)
y0 = -(δx + F(3)) * p_4
s0, c0 = sincos(y0)
cs = ntuple(8) do i
y = (δx + 4 - i) * p_4
# Improve precision of Lanczos OpenCV4 #451, avoid NaN
if iszero(y)
y = eps(oneunit(y))/8
y = δx + 4 - i
if abs(y) >= F(1e-6)
y *= p_4
(F(l4_2d_cs[i, 1]) * s0 + F(l4_2d_cs[i, 2]) * c0) / y^2
else
F(1e30)
end
# Numerator is the sin subtraction identity
# It is equivalent to the following
# f(δx,i) = sin( π/4*( 5*(i-1)-δx-3 ) )
(l4_2d_cs[i, 1] * s0 + l4_2d_cs[i, 2] * c0) / y^2
end
sum_cs = sum(cs)
normed_cs = ntuple(i -> cs[i] / sum_cs, Val(8))
Expand Down
Loading