1.0.1 - RidgeSlide sub-brand in the name
- Display name is now "GPS Road Builder (RidgeSlide)" to link the
plugin to
its RidgeSlide engine while staying discoverable; the window title and
header match. The package folder and settings keys are unchanged
1.0.0 - Sprint 13: RidgeSlide boundary and first production release
- RidgeSlide carved into a clear, QGIS-agnostic core: renamed
core/slide ->
core/ridgeslide (files refine.py, kernel.py) with a small public API
façade (RidgeSlide / RidgeSlideConfig / RidgeSlideResult) so the
method can
be reused as a library; the plugin pipeline stays a client
- The RidgeSlide core is MIT-licensed (core/ridgeslide/LICENSE +
NOTICE, with
the paulmach/slide + Guo 2020 lineage); the plugin stays GPLv3
(compatible)
- Added CITATION.cff and AUTHORS; documented the dual licensing
- A guard test enforces that the algorithm core imports no qgis/PyQt
- First production release (1.0.0)
0.17.0 - Sprint 12: topology and an accurate preset
- Break at crossings (Post-process): split two edges that cross
geometrically but share no node, inserting a real junction (GRASS
v.clean break) — improves connectivity of visually crossing but broken
roads. Optional (creates a false junction at bridges/overpasses)
- Junction consolidation (Post-process): merge clusters of nearby
junction
nodes into one, cleaning the bushy intersections from 8-connectivity
- New preset "Sparse / RidgeSlide (accurate)": finer cell/resample,
less
smoothing and micro-loop cleanup for geometry closer to the reference
(slower — numba recommended)
0.16.0 - Sprint 11: UI polish, RidgeSlide branding, QGIS 3.x
- Window: the minimize button now works (WindowMinimizeButtonHint)
- Tabs: reverted the bold font that clipped the active tab's last
letter;
selection is shown by colour only, with a robust tab width (no
clipping)
- About dialog redesigned as a styled card (like Support),
highlighting the
RidgeSlide method and its authorship
- Support dialog: removed the "Maybe later" button (donate or close
via X)
- Language switcher shows real flag icons (Windows does not render
flag
emoji)
- UI now says "RidgeSlide" everywhere instead of bare "Slide" (tab,
method,
backend, presets, tooltips); internal setting values are unchanged
- Compatibility: documented support for the whole QGIS 3.x branch
(tested on
3.44 "Solothurn"); qgisMaximumVersion=3.99 keeps 4.x/Qt6 out until it
is
separately supported
0.15.0 - Sprint 10: scenario presets
- New one-click presets beyond forestry, since the engine is
method-agnostic: "Dense tracks (running / cycling)", "Urban vehicles /
logistics", "OSM traces (heterogeneous)" and "Sparse vessels / AIS"
- Each encodes assumptions about the data (cell size, resample step,
gap
segmentation, threshold mode) as a starting point — expect to tune
- Dense/urban/OSM use the RidgeSlide method; urban and OSM use the
robust
percentile threshold; AIS uses coarse scale and wide segmentation gaps
0.14.2 - More connectivity + post-processing logging
- Directional bridge: connect dead-ends that face each other (their
road
continuations point at one another) even beyond the normal radius
- Component stitching: join the nearest nodes of separate components
into one
network (Kruskal over k-nearest neighbours), up to a max distance
- Post-processing runs now log their full settings and result counts
(bridged/snapped/faced/stitched/dropped) to the file log and the run
manifest, like build runs
0.14.1 - Node-to-edge gap bridging (post-processing connectivity)
- "Bridge gaps" now also snaps a dangling end to the nearest point on
a
neighbouring edge (a T-gap), splitting that edge — not only
end-to-end.
This is the common road gap that end-to-end bridging could not close,
so
post-processing barely changed the graph before. The Post-process log
now
reports bridged/snapped/components-dropped counts
0.14.0 - Sprint 9: new input sources
- Load input from a project vector layer (points or lines) or a
GPX/KML/SHP
file, not only CSV/XLSX. Vector inputs need no coordinate columns —
the
geometry provides them; device/time are auto-detected from field names
or
synthesized; line layers treat each line as one track; GPX uses its
track_points sub-layer
- Data tab: an "Input source" selector (files / project layer / vector
file)
0.13.1 - RidgeSlide authorship
- Documentation states throughout that RidgeSlide was conceived,
generalized,
developed and completed by Alexander Kobyakov (README, ALGORITHMS,
USER_GUIDE, metadata)
0.13.0 - Sprint 8: product (docs, RidgeSlide, UI fixes)
- The custom Slide reimplementation is now named RidgeSlide (roads =
ridges
of the density surface); metadata/README/ALGORITHMS updated
accordingly
- Marked production-ready (experimental=False); positioning
generalized
beyond forestry (logistics, sports, off-road, vessels, OSM, ...)
- New docs at repo root: rewritten README, USER_GUIDE.md,
ALGORITHMS.md
- UI fixes: the active tab no longer clips its last letter (bold
applied to
all tab states); the AOI group on the Data tab is now visible without
scrolling; the build script excludes test_temp/memory
0.12.1 - Sparse/Slide threshold fix (per-tile Otsu blow-out)
- The ui5 run showed per-tile Otsu thresholds ranging 0.03..374: tiles
with
a town/yard got a huge threshold and dropped almost all roads (sparse
result). The "Sparse / Slide" preset now uses the percentile threshold
(scale-adaptive per tile) instead of Otsu, and a slightly finer cell
(10 m)
- Tip: lower the threshold percentile for a denser network
0.12.0 - Sprint 7: KDE quality and device attributes
- KDE quality (WS-KDE): percentile threshold mode (robust on sparse
data
where Otsu cuts above roads); KDE now uses all points incl. single
fixes;
mask hole-filling and micro-loop removal kill the skeleton
"crossings";
medial-axis (centerline) skeletonizer option; the sparse KDE preset
uses
these by default
- Device attributes (WS-Dev): n_devices per edge (unique GPS devices
near
the edge), written to the layer, GeoJSON and GraphML
- WS-Perf: resuming from a checkpoint already skips the input re-read
(Sprint 5); parallel tiles (A5) intentionally deferred — numba already
parallelizes Slide across cores and a process pool is unsafe in QGIS
0.11.4 - Drop remaining "FGIS LK" mentions
- Removed "ФГИС ЛК" from the KDE method label and its tooltip (kept
only the
generic sparse-data description)
0.11.3 - Preset naming fix
- Sparse presets renamed: "Разреженные / Slide (рекоменд.)" and
"Разреженные / KDE (эксперим.)" — dropped "ФГИС ЛК" from the KDE
preset
name and unified the spelling (разреженные) across the UI
0.11.2 - Backend clarity (numba vs numpy)
- The log now records the resolved Slide backend (setting -> using) so
it is
clear whether numba or numpy actually ran
- The numpy backend option is relabelled "numpy (slow, reference)" and
the
Slide-backend tooltip explains that installing numba on the
Dependencies
tab does not select it — pick "Auto" or "numba"
0.11.1 - Sparse/Slide performance fix
- "Sparse / Slide" preset: numba backend + coarse resample (K=50, gap
700)
so sparse points no longer explode into tens of millions (was ~83M /
225
tiles); coarser cell for fewer tiles
- Guard: Slide on the numpy backend with >15M points now aborts early
with
guidance instead of grinding for hours; a warning is logged when numba
is
available but numpy is selected
- Log now shows the resample point-count inflation
(before/after/ratio)
0.11.0 - Sprint 6: post-processing and connectivity
- Connectivity (WS-Conn): bridge dangling ends within a radius, drop
small
components, keep only the largest network; optional morphological mask
closing for the Slide method too (core/graph/connect.py)
- New "Post-process" tab (WS-Post): improve an already-built graph (or
a
chosen line layer) without a full rebuild — bridge gaps, smooth, clean
components; shared orchestrator core/graph/postops.py
- Area of interest (WS-AOI): clip input points by a polygon from a
file or a
project layer before processing (removes points in the sea / out of
zone)
- Connectivity/AOI settings also apply during a normal build
0.10.0 - Sprint 5: transparency and a working result
- Logging for parameter tuning (WS-L): the run now logs the full
resolved
settings and per-stage metrics (per-tile binarization threshold,
edges/
nodes before/after merge and cleanup) with per-stage durations; a
compact
per-run manifest is appended to gps_road_builder_runs.jsonl; log
rotation
raised to 5 MB; resuming from a checkpoint no longer re-reads the
input
- New "Sparse / Slide (recommended)" preset — Slide builds a connected
network on sparse FGIS LK data; the KDE preset is marked experimental
- Polyline smoothing (WS-Smooth): optional Chaikin smoothing removes
the
raster-skeleton stair-steps (Graph tab, keeps edge ends fixed)
- UI fixes (WS-U): the tab bar no longer scrolls away (scrolling moved
inside each tab) and tab titles are no longer truncated (+ tab
tooltips)
- Presets now apply to all parameter tabs, not just Density/Graph
0.9.0 - Sprint 4 (WS-G): quality and attributes
- Graph post-cleaning against fragmentation: pass-through (degree-2)
nodes
are merged into single polyline edges (always on); optional removal of
short dangling spurs (Graph tab, "remove dangling spurs shorter than")
- Edge attributes: road_class (main/improved/ordinary/winter by
traversal
frequency quantiles) and a "reconstructed" flag (inferred segment with
no
track coverage); written to the layer, GeoJSON and GraphML
- REB / anti-spoofing preprocessing filter (Preprocess tab): drops
teleports
(>120 km/h), jumps >200 m in <60 s and A->B->C spikes
- Overview histograms use log-spaced bins with range labels, so skewed
frequency/length distributions are readable instead of one dominant
bin
0.8.1 - WS-E completion: tooltips and method-aware fields
- Tooltips on every parameter of the Preprocess, Density/Slide, Graph
and
Scale tabs (plain-language hints with recommendations)
- Fields irrelevant to the chosen method are greyed out: KDE disables
resample, Slide backend, sigma1/sigma2, kernel sharpness and Slide
loops;
Slide disables KDE radius/buffer
0.8.0 - Big update, Sprint 3: tester convenience
- Checkpoints: cleaned points and projected tracks are saved to a
chosen
cache folder; a run can start from a saved stage or stop after a
stage, so
you re-tune later steps without re-running the expensive early ones
- Presets can be saved to and loaded from JSON files (not only the
built-ins)
- Output tab: cache folder, "start from stage" and "stop after stage"
0.7.0 - Big update, Sprint 2: sparse-data method
- New "KDE + buffer" method for sparse GPS logging data (FGIS LK: a
fix every
15 min / 5 km) — kernel density on points, gap bridging by a buffer,
no
resampling/Slide; the Guo "Density + Slide" method stays for dense
tracks
- Method and backend selection in the UI (Slide backend, skeletonizer,
KDE
radius/buffer); "Sparse data / FGIS LK" preset
0.6.1 - Big update, Sprint 1: performance on large sparse datasets
- Tile assignment via per-track bbox (was O(tiles*tracks)); single
vectorized
projection; near-duplicate thinning (numba); default cell 5 m + raster
guard; file log fixed, step plan/ETA/available-libraries reporting
0.6.0 - Phase 5: polish
- Export to GeoJSON / GraphML (pure) and SHP / GeoPackage
(QgsVectorFileWriter)
- Rotating file log in the QGIS profile for debugging
- User guide (docs/USER_GUIDE.md), README, cross-platform CI (Linux +
Windows)
- Plugin packaging script excluding data/docs/tests
0.5.0 - Phase 4: functional GUI
- Real tab controls (Data...Output), field presets, tooltips, result
histograms; Build runs the pipeline in a QgsTask and adds the road
layer
styled by traversal frequency; progress, log and cancel
0.4.0 - Phase 3: scalability
- Split-and-merge over overlapping tiles with boundary fixing;
close-node
merge that also cleans junction clusters
0.3.0 - Phase 2: MVP pipeline
- Density surface, improved Slide (numpy reference + numba backend,
verified
against the reference implementation), skeleton -> graph, edge weights
and
(f, l) artifact filtering
0.2.0 - Phase 1: data and CRS
- Robust CSV/XLSX reader (delimiter and datetime auto-detection),
cleaning,
deduplication, segmentation, equidistant resampling; data-centered
Transverse Mercator working CRS
0.1.0 - Phase 0: plugin skeleton and UI chassis
- Tabbed main dialog (header with language selector, Support and
About)
- Dependency installer framework (pip-into-target and
prebuilt-wheels), no
silent auto-install; runs only on explicit user action
- Background task scaffold (QgsTask), settings persistence, RU/EN
interface
- Offline unit tests and CI

-