GameCult/repixelizer

Watch

Fork 0

Meta Vens ad234cd30e Revert bad phase-field tuning pass

2026-04-24 00:13:48 +01:00

4.4 KiB

Raw Blame History

Repixelizer Implementation Plan

Current machine

Repixelizer now has two real reconstruction engines:

phase-field: the default optimizer in src/repixelizer/phase_field.py
tile-graph: the source-owned alternate solver in src/repixelizer/tile_graph.py

The old tray-based optimizer is gone. Good riddance.

The live pipeline is:

source image -> lattice inference -> edge analysis -> phase-field or tile-graph reconstruction -> cleanup -> optional palette fit -> diagnostics

What is working

lattice size and phase inference are still shared and still CUDA-capable
phase-field is now the main optimizer path and produces the best-looking badge result in the repo so far
tile-graph still provides a useful alternate path when literal source ownership matters more than smooth field behavior
compare mode, benchmark mode, diagnostics writing, and tuning all still work after the optimizer cutover
the repo now reports both:
- source_fidelity: agreement with the inferred lattice portrait
- source_structure: visible structural agreement at source size

That second metric exists because the first one was happily slandering the better-looking output.

Current evidence

Phase-field

Pinned badge case:

fixture: tests/fixtures/real/ai-badge-cleaned.png
lattice: 126x126
phase: (0.0, -0.2)
steps: 48

Useful artifacts:

first pinned badge run: artifacts/phase-field-v1-badge-126/
tip-focused follow-up: artifacts/phase-field-v2-badge-126/
structure-metric sanity check: artifacts/phase-field-metric-check/

What we know:

the output preserves the important badge structure better than the deleted optimizer did
the tracked blemish is now narrow and specific: the sword-tip stroke widens too much in one local region
the tracked focus fixture is tests/fixtures/real/ai-badge-tip-focus.json

Tile-graph

Useful artifacts:

full emblem win: artifacts/full-emblem-tile-graph-atomic-v3-cuda/
pinned badge speed/quality checkpoint: artifacts/tile-graph-reducebykey-v1-badge-126/

What we know:

source ownership is real now
large-fixture cold-build cost is still the main pain
the dominant remaining speed problem is connected-component labeling / region extraction, not the parity solver loop

Maps

The repo now keeps the living maps only:

docs/lean-optimizer-algorithm-map.md
docs/tile-graph-algorithm-map.md

If a future pass cannot be explained cleanly against one of those maps, it should not land.

Current priorities

1. Fix the phase-field sword-tip blemish

Goal:

keep the current structural win
stop widening the dark contour near the tapered sword tip

Current hypothesis:

scalar weight nudges are not enough
the field needs better anisotropic behavior near sharp tapered contours
it likely needs to distinguish motion along a stroke from motion across a stroke

What we just learned:

the later d9fa411 phase-field tuning pass was a real regression, not paranoia
that pass strengthened the local edge penalty, added a spacing loss plus upper-spacing clamp, and made edge gating more aggressive
on the pinned badge case, it removed internal linework and changed 2433 / 15876 output cells relative to the original good phase-field run
reverting that tuning pass restores the original good badge result exactly; the fresh fixed run under artifacts/phase-field-regression-fix-badge-126/ is byte-identical to the original artifacts/phase-field-v1-recheck-badge-126/

Tracked fixture:

tests/fixtures/real/ai-badge-tip-focus.json

2. Keep phase-field honest

Rules:

do not reintroduce portrait layers, candidate trays, or solver-stage religions
if a new term lands, it must belong cleanly in the one-field objective
if a result looks better and the metrics disagree, fix the metrics rather than worshipping them

3. Make tile-graph less slow