master
2 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
| 136492b197 |
Fix ally marker detection; expand TargetType icons and pickers
map_vision.py: marker shape test only matched diamonds, so friendly
(rectangle) markers could never be detected regardless of color match.
diamonds() now takes a per-side ideal shape (diamond for hostile, full
rectangle for friendly) with fill-ratio bands measured off real markers.
Merged tests/fixtures/map_shots/more/ into the main fixture set: 2
screenshots that solve fine (now 15.png/16.png, with hand-transcribed
ground truth) and 8 that are too low native resolution for the label
reader (same class as the existing 12.png) into too_hard/ as
17.png-24.png, with an explanatory README entry. Updated map_vision.py's
docstring numbers (9/12 solve, 104/131 points correct) to match.
TargetType: expanded from 11 to 44 members to cover every icon in
assets/icons/targets/{enemy,friendly}/, including 7 friendly-only types
(King, Police, General, Hospital, Fort, Civil-Military, Mechanized
Anti-Tank) with no enemy equivalent. UNKNOWN/ENEMY stay icon-less by
design (both are literal words the game's OCR'd text uses, confirmed via
ocr.py's _TYPE_BY_SHORT, so neither can be dropped without breaking real
parsing) and draw the same plain-dot fallback the map itself uses.
icons.py: collapsed the icon lookup into one canonical table
(_TARGET_ICON: TargetType -> (enemy_basename, friendly_basename), one
explicit row per type) instead of a basename table plus two exception
dicts layered on top -- with a startup assertion that every TargetType
has a row. Added build_target_type_grid(), an icon-grid picker (icon +
name, same idea as the existing Shell picker) that replaces the old
plain-text dropdown/list everywhere a type is chosen, and only offers
types the given side actually has real art for.
coord_dialog.py: Add/Edit Target and Add Ally now use the icon grid
instead of Adw.ComboRow. Fixed a resulting horizontal-scroll bug (an
unbreakable long word was blowing out cell width) and locked the
coordinate pickers back to 5 columns.
app.py: right-click quick-add now offers Spotter/RP alongside
Target/Ally/Strike, opens a real modal (not a Popover, which turned out
unreliable for a wide multi-row grid) to ask for a type instead of
silently defaulting to UNKNOWN, and doesn't repeat the coordinate on
every row. The type grid listens for "clicked" rather than "toggled" --
a grouped ToggleButton doesn't emit "toggled" when you click the one
that's already active, which meant confirming the pre-selected default
type silently did nothing.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
|||
| ff3b89c41b |
Add the map-table vision pipeline
Reads a screenshot of the in-game map table and recovers where on the board it is looking. The scene is a flat table under a perspective camera, so a single homography describes grid<->screen exactly; that is fitted from line evidence (LSD segments, vanishing-point RANSAC, 1-D lattice fits). Line evidence alone cannot finish the job: the 1km cells and the 100m subgrid are locally identical, so it fixes neither scale, axis assignment, axis direction nor phase. Those come from the cell labels -- and the labels are never *detected*, they are correlated where the grid says they must be (9% in from a cell's left edge, 6% down from its top). Blob detection on aerial-photo terrain finds texture, not glyphs; correlating a known template at a known place has no such failure mode. The same score then ranks the geometry hypotheses, since a wrong lattice puts the crop where no label is, so one number resolves scale, axis assignment, direction, phase and anchor together. Markers are found by hostile/friendly colour plus an IoU test against an ideal inscribed diamond, which cut a red-lit shot from 43 false positives to 2 while preserving every hand-verified count. Unit-type classification is present but not yet reliable, and returns unknown rather than guessing. Measured over the fixture set: 7 of 10 solve, each with 100% of its ground-truth points in the correct cell (85/112 overall), residual spread 0.005-0.033 cells. The other three reject cleanly; none has ever produced a plausible-but-wrong grid. Across 122 typewriter screenshots solve() accepted none, which is what makes clipboard routing safe. Fixtures: 10 map shots with hand-transcribed ground truth, plus 10 typewriter shots spanning 262-5366px for routing and future OCR tests. Map shots wider than 2400px (the pipeline's own maximum working width) were downscaled with their coordinates rescaled to match; re-measured afterwards, the results are identical. Typewriter shots stay at native resolution because OCR needs the text legible. Drops docs/map_vision_plan.md and its WIP prototype: the design now lives in the module docstring, next to the code it describes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
