Repository navigation
feat(render): recover crisp playable surface boundaries - #43
Merged
Merged
Conversation
Task 1 (green step) of the P2a crisp-surfaces plan. The green->tan boundary was the SUM of three soft layers (soft splat base, ~1.8m collar dilation, low-freq aerial tint); sharpening any one didn't help (Task-0 gate finding). Fix: composite the surface base colour IN-SHADER, gated by a crisp fwidth mask, overriding all three AT the boundary. turf.js (cache key v32 -> v33): - uMaskRaw (RAW, unblurred mask) + uPalGreenA (linear palette green) uniforms. - gCrisp = smoothstep(0.5-fwidth(gRaw), 0.5+fwidth(gRaw), gRaw) — collapses the 0.45 m/px mask ramp to a ~1px mow line at the polygon edge. - baseCol = mix(diffuseColor, uPalGreenA, gCrisp): crisp green base override. - gEdge = gCrisp so the v29 checker/contour/sheen ride the crisp edge (they still apply INSIDE where gCrisp==1 — verified they survive). - gVic (~4m green-vicinity dilation) suppresses the low-freq aerial tint near greens so the blurred green oval can't smear a soft halo across the crisp line. - collar dilation tightened 1.8m -> 0.6m (gBlur now feeds only fr, not the character); a proper offset fescue collar is Task 2. scene.js: - _paintMask gains a blurPx param; a RAW (blurPx 0) packed mask is built and wired through _turfInputs (base + HD-patch share it) alongside the linear palette. - green splat blur 1.0 -> 0.35 so the splat doesn't bleed green past the polygon where gCrisp==0 falls back to it. fixtures: green_close repointed 138,-289 (a TEE — the green-look there is aerial photo, not an OSM green, so gCrisp is correctly 0) -> 193,-263 (a real OSM green). Verified before/after on the real green (193,-263): v32 soft ~2m airbrush -> v33 crisp mow line; v29 checker/contour intact; P1b tan fescue unregressed. Non-green fragments byte-unchanged (gCrisp/gVic == 0 away from greens). 301/301 tests green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fairway/rough step of Task 1. The fairway's soft cue is the STRIPE FADE (the splat fairway->rough colour is already near-crisp at ~0.5m and both sides are close tans/olives). So: - mCrisp = fwidth-AA on the RAW mown mask (.r) — one shared raw-mask sample now fetches both .r (mown) and .g (green). - the mow-stripe gate keys off max(mCrisp, cls.r) instead of the soft `m`, so stripes stop crisply at the OSM mow line (NDVI coverage cls.r preserved; its feather is reconciled in Task 3). Deliberately did NOT override the fairway BASE colour: forcing the dry-olive fairwayA over the whole mown gate recolours the fairway and reads near-black on shaded slopes (verified regression) — that's scope creep past "crisp edges". The green keeps its full base override; the fairway gets the crisp stripe edge only. Verified on Chambers: fairway bright + striped (no darkening), stripes terminate on the mow line; green edge + P1b tan unregressed. 301/301 tests green (v33). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…band Bunker step of Task 1. The blurred OSM bunker mask left a soft desaturated "sand halo" band around each bunker where the sand->grass gate ramped. Crisp it: - bCrisp = smoothstep(0.5-fwidth(bRaw), 0.5+fwidth(bRaw), bRaw) on the OSM bunker mask, so the sand gate bm = max(bCrisp, cls.b*(1-m)) — a ~1px sand->grass line. - The NDVI-detected sand (cls.b) union is unchanged; its feather is reconciled in Task 3. Verified on Chambers: the near/play bunkers read with crisp scalloped edges, the soft halo band is gone; grass/green/fairway unregressed. 301/301 tests green (v33). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replaces the placeholder fr (a faint 16% darkening) with a distinct collar/fringe: a ~0.8m apron ring just OUTSIDE the crisp putting-surface edge, coloured a touch lighter + warmer than the green, with a crisp INNER mow line and a soft OUTER edge into the rough, distance-faded (play/mid feature — far photo owns past ~60m). turf.js (cache key v33 -> v34): - collarBand = clamp(gDil,0,1) * (1-gCrisp) * gDistFade, where gDil is an 8-tap dilation of the RAW green mask at 0.8m (soft outer edge); (1-gCrisp) = crisp inner. - 3-way base-colour stack: rough(splat) -> collarCol apron -> uPalGreenA putting surface. Compositing in baseCol means the collar gets the same grain/warm-mix/ sun-rake/desat as everything else — reads as real mown grass, not a decal. - collarCol = uPalGreenA * vec3(1.25,1.15,0.90) (tunable apron knobs); no new palette plumbing (derived in-shader from the existing green uniform). - Removed the old blurred gBlur 8-tap block + fr; stripes now suppressed on the collar via collarBand (was fr); dropped the fr darkening line. Net cleanup. Approximate collar (mask dilation), per the Task-0 gate's sanctioned fallback — no per-green/whole-course SDF. All terrain meshes share the material so the collar reaches the ring outside every green. Verified live on Chambers (links), Sawgrass (parkland, green-on-green), St Andrews (rough-OSM, 88 greens): a distinct lighter-green apron ring, crisp inner + soft outer, no sticker read at mid, no egregious artifact; v29 checker/contour + Task-1 crisp edge + P1b tan unregressed. Non-green fragments byte-unchanged (collarBand==0 away from greens). 301/301 tests green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…dge (v35) The crisp OSM edges (Task 1) stacked on top of the FEATHERED NDVI classmap union (cls.r mown, cls.b sand, ~feathered coverage) produced a DOUBLE edge: a crisp OSM line + a soft NDVI halo on the SAME boundary — most visible as a pale desaturated "sand halo" ringing every OSM bunker. Fix (turf.js, cache key v34 -> v35): - osmNear = a 4-tap MAX dilation of the RAW OSM mask (mown .r OR bunker .b) over ~5m — a per-fragment "OSM authored a surface nearby" signal. - In macroPre, before the mown union: cls.r *= 1-osmNear; cls.b *= 1-osmNear. Where OSM already authored the boundary the crisp OSM edge OWNS it (NDVI feather removed); NDVI survives only in genuine OSM gaps (osmNear==0) — its coverage role is preserved. Verified on Chambers (bunker sand halo GONE, sand->grass edges now clean/crisp; fairway unchanged — its NDVI mown was redundant with OSM; overview clean, no coverage holes/rings) + Sawgrass (bunkers clean, green complex + water healthy). No over-suppression. 301/301 tests. (P0a dot-screen + macro seam are still Task 4.) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… + threshold (v36) The runtime NDVI classmap is per-pixel scatter (false positives NDVI misreads as mown/sand); unioned raw into the turf it rendered as a "dot-screen" checker/halftone stipple across the ground. Confirmed by a live classmap→black swap (stipple vanished). Fix (no data regen): - scene.js: blur the classmap into SMOOTH coverage at LOAD (a 4px canvas blur in the loader callback) — isolated noise averages toward 0, contiguous real regions stay high. - turf.js (cache key v35 -> v36): threshold the sampled classmap `cls.r/b = smoothstep(0.45, 0.75, ...)` so only CONFIDENT coverage adds surface; the blurred low-amplitude noise is dropped. (Task-3 osmNear suppression kept.) Verified on Chambers: the near/mid dot-screen is GONE (matches a classmap-off reference); fairways/greens/stripes intact (stripes are OSM-driven; only redundant/noisy NDVI was removed). 301/301 tests. NB the HD-patch rectangular seam is a SEPARATE issue (persists with the classmap off) — next commit. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
1 participant
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.
Outcome
Recover and correct the reviewed P2a surface-readability work on the current SP-00 base.
Changes
Exact evidence
Scope notes