Dominik Roth
ff3b89c41b
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>
1.6 KiB
Test fixtures
Real screenshots of the game, used as regression data. They are the game
author's work, not covered by this repo's MIT license (see /LICENSE).
map_shots/ + map_shots_gt.json
The evaluation set for the map-grid solver (src/fenigma/map_vision.py), scored
by tools/eval_map_vision.py. The JSON holds hand-transcribed cell labels at
native pixel positions; its own header comment explains the format and the
9%/6% label-padding constant.
Shots wider than 2400px were downscaled to 2400px, and their ground-truth
coordinates rescaled with them. 2400 is map_vision.RETRY_WORK_W, the widest
the pipeline ever works at, so nothing the code can actually read was lost.
Measured after the downscale: the same 7 of 10 solve, 100% of their points land
in the correct cell, residual spread unchanged.
too_hard/ holds shots that are permanent rejections; see its own README.
writer_shots/
Typewriter/field-log screenshots. Two uses:
- Measuring false positives in the map-vs-text routing gate
(
map_vision.looks_like_map/solve). Over the full 122-shot set the cheap gate false-positived on 6% andsolve()accepted none. - OCR regression material for
ocr.py.
Kept at NATIVE resolution deliberately: the routing gate only ever sees 1500px, but OCR needs the text legible, so these must not be downscaled.
Ten shots are committed, chosen to span the capture-scale range (262px to 5366px wide) since scale is what both the gate and OCR are sensitive to. The false-positive numbers above were measured on all 122; this subset is a regression guard, not the measurement.