Spirula Studio meshes the splats inside the room (clean.py --outside); the result stands upright with the
floor at height 0 – Z up for OBJ/PLY/STL, Y up for GLB. Texture when the training images are still there.
Tested on real scenes (kitchen without images: colours per corner; textured on one with images) and with a
fake spirula mesh in the tests (579 passing). Not done yet: look at the dialog in the browser, changelog,
README, release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From issues #1 and #2:
- Unknown settings in config.toml (e.g. splat-transform instead of splat_transform) show up in the
setup hints and in install.sh, with the right spelling.
- install.sh writes the full path of splat-transform into a new config or prints the line for an
existing one: started from the desktop, the app does not see the shell's PATH.
- The setup guide's npm --prefix instructions match the path they tell you to configure.
- A pipeline not pinned below PyAV 19 is spotted before a run; a run failing on it says why.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Dialog: for the Frame (one size, with measured fps) or for the PC (all splats); clean-up; optionally
the same without; a preview of what will be made and what it replaces; one version per click.
- Names carry the clean-up strengths; As template takes a version's settings into the form.
- A cleaned PC version keeps its ply; /api/open opens a version in Spirula Studio.
- Versions unfold under their scene or tour, with Spirula, Frame, template and delete.
- The PC sizes 3/5/8 M are gone; measured frame rates updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- clean.py: the ceiling is the highest strong line where walls meet it (a cupboard top lies lower).
- The API takes the clean-up as a flat string (ceiling=0.5,haze=0.3), like every other field.
- tests/test_clean.py: a synthetic living room with thin ceiling, veil, cupboard top, bird cage,
haze, a hole in the floor and things beyond the walls – each told apart by its colour.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found moving the runs to an NTFS disk: the tour "Rundgang 12.10." (folder Rundgang-12.10.) was not copied.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The watcher reports the next part (or putting the parts together) as running in the tick that
starts it: no false standby notice between parts.
- Deleting an unfinished part deletes its piece of video in runs/.parts too.
- combine.sh keeps the floor plan's labels (plan.json).
- Tests: the chain through the watcher's tick (twice), one standby notice per chain, the own lock
released before suspending, rooms of the same name, view-dependent colour matched, the preview plan
of a home named by rooms; notifications are awaited instead of checked too early.
- 0.3.0: version, changelog, READMEs without the preview notice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A checkbox on the running job: when everything is finished (parts, putting them together, copying),
a minute's notice, then systemctl suspend; cancelling a job switches it off; a refused standby is
reported. Our own lock is released first (it kept the PC up in a first try by hand).
- One sleep lock of the server spans the seconds between two jobs (parts of a long recording).
- A renamed room's floor plan and previews are drawn again with the new name.
- Rooms of the same name: "Kitchen (2)"; the combine dialog and job card speak of a room's recordings.
- README: filming tips for ceilings and light; 0.3 notes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A place only goes to a recording that really has something there (a quarter of the richest one's splat
opacity, over the place and its neighbours). Colours of all recordings are matched per channel from the
places they share (--colour match|keep, COLOUR for combine.sh).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each part gets up to 500 image pairs; parts are cut losslessly with the original time stamps, converted
one after the other and put together as a room. The original stays (asked once more before deleting).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A group can be a room (several recordings of one room); any other group may hold scenes and rooms and
is put together from the recordings, each in its room. Recordings of one room no longer count as
walking in to each other, which also hid a real walk-in from another room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Documents/GaussianXR is left alone by Steam updates. Old copies in the app's own Scenes folder
move over when copied again, or all at once from the On the Frame tab.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tab lists everything in Gaussian XR's gallery on the Frame with its preview image and what it
is (scene, tour, version, or not from Splat Workshop) and takes any of it off the Frame. Scenes are
copied straight into the gallery; copies left in Documents/GaussianSplats go along with their scene.
The tour dialog offers 0.75–2 M or an own size. 250 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
For the Frame 0.75 / 1 / 1.5 / 2 M, for the PC all / 3 / 5 / 8 M, plus an own size; with a scale
also walkable in the original's size. Every version is a complete scene in <run>/versions/<key>/
and gets its own scene folder on the Frame, so e.g. 0.75 M and 2 M can be compared in the gallery.
The tour dialog offers 0.75–2 M or an own size for its Frame version. Measured frame rates are
shown as information only. 14 new tests (240 in all).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Floor plan, sizes, scale bar and measuring tool in the original's size; the floor plan measures
walls and floors in the model's scale. Optionally Gaussian XR's worldScale is set when copying to
the Frame (not tested on the Frame yet). The floor plan no longer fails without walls.
13 new tests (226 in all).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Fakes for kdialog, zenity and xdg-open; every test catches shutdown/os._exit
- Broken download, frame folder found on rename, nested values in non-text fields,
Start over from the cancelled runs, the watcher's notification and tidying up
(watcher split into watch_once for that)
- Fake pipeline refuses to continue an unfinished extraction, like the real one
213 tests, every mutation of both review rounds is caught.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
206 tests instead of 57; every surviving mutation of the review is caught now. New fakes for
ffprobe, ssh/scp (the Frame as a folder) and notify-send; synthetic rooms with tracking noise,
scale error, turned splats, tracking turn and an adjustable walk-in. Fixed: wrongly typed
members/key/align/partition gave a 500 or were stored.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
57 tests without GPU, recordings or Frame: fake pipeline, fake splat-transform and trash, and
synthetic rooms with a known error for putting rooms together. Found on the way: after cancelling,
the job counted as finished when systemd-inhibit ended, while splat.sh was still putting the
previous scene back – pid_alive now looks at the whole process group.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
