Generate clean 2D game sprites & animation atlases — component-row pipeline: state rows, alpha cleanup, frame extraction, runtime atlases. Codex/Claude skill.
fix(video): bootstrap diagnostics and remove redundant return guards
Soohong Kim5c8e1aa0bfa3
diff --git a/scripts/diagnose_video_loop.py b/scripts/diagnose_video_loop.pyindex 5b2edd6..daffbc2 100644--- a/scripts/diagnose_video_loop.py+++ b/scripts/diagnose_video_loop.py@@ -1,18 +1,22 @@# SPDX-License-Identifier: Apache-2.0"""Measure existing keyed frames without changing loop selection or acceptance.-Run from the repository with PYTHONPATH=. using its Python environment. Outputs+Run using the repository's Python environment. Outputsare private diagnostic artifacts: keep them outside the repository when inputsare private. Frame indices are zero-based. Wide-window candidates are evidence,not accepted animations: low seam alone cannot establish a complete action."""from __future__ import annotations+import sys+from pathlib import Path++sys.path.insert(0, str(Path(__file__).resolve().parents[1]))+import argparseimport hashlibimport jsonimport math-from pathlib import Pathfrom PIL import Image, ImageDrawdiff --git a/sprite_gen/video/loop.py b/sprite_gen/video/loop.pyindex 0b8e752..4c0048f 100644--- a/sprite_gen/video/loop.py+++ b/sprite_gen/video/loop.py@@ -312,7 +312,6 @@ def detect_one_shot(D: np.ndarray, *, min_len: int, max_len: int, frame_mass: npn = len(D)adjacent = np.diag(D, 1)candidates = []- rejected_edges = 0max_moved = 0.0for start in range(n - ONE_SHOT_MIN_LEN + 1):for end in range(start + ONE_SHOT_MIN_LEN - 1, min(n, start + max_len)):@@ -344,16 +343,10 @@ def detect_one_shot(D: np.ndarray, *, min_len: int, max_len: int, frame_mass: npa -= 1while b + 1 < n and active[b + 1]:b += 1- if a == 0 or b == n - 1:- rejected_edges += 1- continueif b - a + 1 < ONE_SHOT_MIN_ACTIVE:continueif not (start <= a - ONE_SHOT_PAD and end >= b + ONE_SHOT_PAD):continue- # An endpoint inside another excursion is not a rest witness.- if active[start] or active[end]:- continueinner = float(adjacent[start:end].mean())if inner <= 0 or departure < ONE_SHOT_MIN_COHERENCE * inner:continue@@ -374,15 +367,15 @@ def detect_one_shot(D: np.ndarray, *, min_len: int, max_len: int, frame_mass: npraise CycleSelectionError(f"video-loop: no complete one-shot return — the clip never leaves its rest pose "f"and returns with observed endpoints (moved {max_moved:.2f}, "- f"need {ONE_SHOT_MIN_MOVED}; boundary excursions refused: {rejected_edges})",- {"kind": "one-shot", "candidates": [], "boundary_excursions_refused": rejected_edges},+ f"need {ONE_SHOT_MIN_MOVED})",+ {"kind": "one-shot", "candidates": []},)# Keep the full strongest action, then minimise excess rest. Do not choose a# tiny low-seam twitch just because its endpoints happen to match exactly.strongest = max(c["departure"] for c in candidates)eligible = [c for c in candidates if c["departure"] >= strongest * 0.95]best = min(eligible, key=lambda c: (c["length"], c["ratio"], c["start"]))- return {**best, "candidates": candidates, "boundary_excursions_refused": rejected_edges}+ return {**best, "candidates": candidates}def fixed_cycle(D: np.ndarray, *, start: int, length: int) -> dict[str, Any]: