Repository navigation
Feat/dither algorithms - #534
Merged
Merged
Conversation
knipknap
requested changes
Oct 10, 2026
knipknap
left a comment
knipknap
left a comment
Contributor
There was a problem hiding this comment.
Thanks again for this PR! The code looks good to me; except for the merge conflicts resulting from your previous PR.
Note that in general I do not usually update .po files as part of new features; I make a single .po update ahead of each release. This avoids merge conflicts like these.
The step settings pages pushed raw attribute values into the varset rows. For enum-backed attributes such as EngraveStep.dither_algorithm the combo adapter got "DitherAlgorithm.BAYER4", matched no entry and kept showing the first one (Floyd Steinberg), both when the dialog opened and after undo or an external change. Serialize the values with Step.recipe_value(), the same way recipes store them, before handing them to the rows.
Extend DitherAlgorithm with Atkinson, Stucki, Jarvis-Judice-Ninke, Sierra (3-row, 2-row, Lite) and Burkes error diffusion, a clustered-dot Newsprint screen and a round-dot Halftone screen. - apply_error_diffusion() takes any forward kernel and an optional serpentine scan. The error within a row is propagated per pixel, the error for the following rows with one vectorised add per kernel tap, which dithers a 2.5 MP image in about 0.6 s. Like raygeo's Floyd-Steinberg it works in linear light. - apply_halftone() lays a cosine spot screen out in millimetres, so the dots stay round on the non-square raster pixels, and ranks the spot values so the dot area follows the darkness. - Newsprint tiles a 45 degree clustered-dot matrix whose cells scale with the minimum feature size and stay square in millimetres. Floyd-Steinberg without serpentine and the Bayer screens keep using raygeo, so existing jobs produce identical output. Refs barebaric#514
The Engrave step's dither algorithm list now offers every DitherAlgorithm. New step settings, saved with the document, recipes and cache params: - Serpentine: alternate the scan direction per row, shown for the error-diffusion algorithms. - Halftone Cell Size and Halftone Angle, shown for Halftone. preprocess_raster_image() forwards them together with the Y resolution, so the screens can keep their cells square in millimetres. apply_import_settings() accepts the depth mode, threshold and dither settings so importers can set them. Refs barebaric#514
CutSetting_Img layers carried their image mode in ditherMode, which the importer ignored, so every imported photo fell back to Variable Power. Map it onto the Engrave step: dithered, atkinson, stucki, jarvis, ordered, newsprint and halftone become the matching dither algorithm (halftone with its cellsPerInch and halftoneAngle), threshold becomes Constant Power and grayscale Variable Power. Sketch has no counterpart and keeps the default. Refs barebaric#514
switch87
force-pushed
the
feat/dither-algorithms
branch
from
October 10, 2026 13:03
7a84c70 to
1c1992f
Compare
Contributor
Author
|
Addressed — rebased onto current main and resolved the conflicts with the previous PR. This PR no longer changes any translation catalogs (zero .po/.pot diff): the new strings live in the source only and the catalog pass stays yours, ahead of each release. The other open PRs (#532, #535, #536, #538, #539, #540) got the same treatment, so none of them should conflict with the release translation pass anymore. |
The error-diffusion kernels, the serpentine scan and the halftone and newsprint screens now live in raygeo (barebaric/raygeo#55, released as 1.62.0) instead of being reimplemented here. rayforge/image/dither keeps the DitherAlgorithm catalogue and dispatches to raygeo, translating the enum values to raygeo's kernel names; the Python-side kernels, spot function and screen builders are gone. Bump the raygeo pin to ==1.62.0 in pixi.toml, requirements.txt and debian/requirements-bundle.txt, and relock pixi.lock.
Contributor
|
Thanks for updating the PR! I made a bit of a change on top of it: I migrated the actual dither implementation to Raygeo for consistency with the existing algorithms. Thanks again! |
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Added
Engrave: more dithering algorithms for photos. The error-diffusion
kernels Atkinson, Stucki, Jarvis-Judice-Ninke, Sierra (3-row,
2-row, Lite) and Burkes, with an optional serpentine scan, plus a
clustered-dot Newsprint screen and a round-dot Halftone screen with
adjustable cell size and angle. LightBurn imports keep the image
layer's dither mode (#514)
Fixed
Step settings: choice rows backed by an enum (such as the Engrave
step's dithering algorithm) now show the step's actual value when
the dialog opens and after undo, instead of the first entry