Add segmented fallback for unmatched beats

README.md

@@ -241,6 +241,13 @@ Insel; der HTML-Report setzt diese Source-Segmente frame-lockend zusammen und
 füllt nur echte Zwischenlücken mit Schwarz. Dadurch können per Blende verbundene
 Trailer-Einstellungen innerhalb eines Beats getrennt gematcht werden, ohne die
 globale Scene Detection aggressiver oder beat-spezifisch zu kuratieren.
+Falls ein kompletter Beat keinen belastbaren Einzelclip ergibt, versucht der
+Matcher dieselbe Segmentlogik automatisch als Fallback: sichtbare Inseln werden
+einzeln global gesucht und anschließend wieder zu einem Beat-Ergebnis
+zusammengesetzt. Sehr kurze Inseln dürfen zusätzlich in den Source-Szenen
+benachbarter bereits gematchter Beats lokal nach ihrer Bewegungsphase suchen.
+Das ist weiterhin nur ein allgemeiner Continuity-Anker, kein manueller Override
+für bestimmte Beat-Nummern oder Szenen.
 Zusätzlich werden sehr dunkle, kontrastarme oder noch nicht sauber
 auf-/abgeblendete Referenzframes aus Score, Inhalts-Reranking,
 Phasen-Alignment und Motion-Templates herausgenommen. Blenden sollen bestimmen,

cli.py

@@ -208,6 +208,22 @@ def _normalize_cached_results(beats: list, results: list, cfg) -> list:
     normalized = []
     for result in results:
         beat = beats_by_id.get(result.beat_id)
+        if getattr(result, "segments", ()):
+            segment_duration = sum(max(0.0, float(s.duration_s)) for s in result.segments)
+            weighted_score = (
+                sum(max(0.0, float(s.duration_s)) * float(s.match_score) for s in result.segments)
+                / segment_duration
+                if segment_duration > 0 else result.match_score
+            )
+            if weighted_score < cfg.cv.deep_scan.provisional_match_threshold:
+                continue
+            if beat is not None and beat.duration_s > 0:
+                coverage = segment_duration / beat.duration_s
+                if coverage < cfg.cv.deep_scan.min_duration_coverage:
+                    continue
+            normalized.append(replace(result, match_score=weighted_score))
+            continue
+
         if result.match_score < cfg.cv.deep_scan.provisional_match_threshold:
             continue
 
@@ -482,6 +498,9 @@ def _attach_visual_segments(results: list, beats: list, cfg) -> list:
         if beat is None:
             expanded.append(result)
             continue
+        if getattr(result, "segments", ()):
+            expanded.append(result)
+            continue
 
         islands = _reference_scoreable_segments(beat, cfg)
         if len(islands) <= 1:
@@ -539,6 +558,161 @@ def _attach_visual_segments(results: list, beats: list, cfg) -> list:
     return expanded
 
 
+def _match_unmatched_visual_segments(results: list, beats: list, cached: list, cfg) -> list:
+    """Create segmented provisional matches when a whole beat has no single match."""
+    from dataclasses import replace
+    from src.core.models import MatchResult, MatchSegment
+    from src.cv.frame_extractor import get_video_info
+    from src.cv.global_scan import run_global_scan
+
+    matched_ids = {r.beat_id for r in results}
+    expanded = list(results)
+    try:
+        fps = float(get_video_info(cfg.paths.source_movie)["fps"]) or cfg.export.edl_frame_rate
+    except Exception:
+        fps = cfg.export.edl_frame_rate
+
+    for beat in beats:
+        if beat.beat_id in matched_ids:
+            continue
+
+        islands = _reference_scoreable_segments(beat, cfg)
+        if not islands:
+            continue
+
+        segments: list[MatchSegment] = []
+        for start_s, end_s in islands:
+            segment_beat = replace(
+                beat,
+                start_s=beat.start_s + start_s,
+                end_s=beat.start_s + end_s,
+            )
+            continuity = _continuity_seed_in_points(
+                beat.beat_id,
+                [b if b.beat_id != beat.beat_id else segment_beat for b in beats],
+                cached + expanded,
+                cfg,
+            )
+            segment_matches = run_global_scan(
+                [segment_beat],
+                cfg,
+                seed_in_points=continuity,
+            )
+            if not segment_matches:
+                local_segment = _local_same_scene_segment_match(
+                    segment_beat,
+                    beat,
+                    start_s,
+                    cached + expanded,
+                    cfg,
+                )
+                if local_segment is not None:
+                    segments.append(local_segment)
+                continue
+            seg = segment_matches[0]
+            seg_dur = min(max(0.0, end_s - start_s), max(0.0, seg.duration_s))
+            segments.append(
+                MatchSegment(
+                    trailer_offset_s=start_s,
+                    duration_s=seg_dur,
+                    scene_id=seg.scene_id,
+                    in_point_s=seg.in_point_s,
+                    out_point_s=seg.in_point_s + seg_dur,
+                    match_score=seg.match_score,
+                    is_confirmed=seg.is_confirmed,
+                )
+            )
+
+        if not segments:
+            continue
+
+        first = segments[0]
+        total_segment_duration = sum(max(0.0, s.duration_s) for s in segments)
+        score = (
+            sum(max(0.0, s.duration_s) * s.match_score for s in segments) / total_segment_duration
+            if total_segment_duration > 0 else min(s.match_score for s in segments)
+        )
+        expanded.append(
+            MatchResult(
+                beat_id=beat.beat_id,
+                scene_id=first.scene_id,
+                source_path=cfg.paths.source_movie,
+                in_point_s=first.in_point_s,
+                out_point_s=first.out_point_s,
+                in_point_frame=int(max(0.0, first.in_point_s) * fps),
+                match_score=score,
+                is_confirmed=all(s.is_confirmed for s in segments),
+                segments=tuple(segments),
+            )
+        )
+
+    return expanded
+
+
+def _local_same_scene_segment_match(segment_beat, beat, segment_offset_s: float, cached: list, cfg):
+    """Find a short trailer island inside scenes adjacent to neighbouring beat matches."""
+    from src.core.models import MatchSegment
+    from src.cv.frame_extractor import open_video
+    from src.cv.global_scan import _content_alignment_score, _content_alignment_templates
+
+    scenes = _load_scene_cache_light(cfg)
+    if not scenes:
+        return None
+
+    by_id = {r.beat_id: r for r in cached}
+    scene_ids: list[int] = []
+    for neighbour_id in (beat.beat_id - 1, beat.beat_id + 1):
+        result = by_id.get(neighbour_id)
+        if result is None:
+            continue
+        ids = [getattr(s, "scene_id", result.scene_id) for s in getattr(result, "segments", ())] or [result.scene_id]
+        for scene_id in ids:
+            if scene_id not in scene_ids:
+                scene_ids.append(scene_id)
+
+    if not scene_ids:
+        return None
+
+    templates = _content_alignment_templates(segment_beat, cfg)
+    if not templates:
+        return None
+
+    min_score = min(
+        cfg.cv.deep_scan.provisional_content_threshold * 0.70,
+        cfg.cv.deep_scan.provisional_match_threshold,
+    )
+    step_s = max(1.0 / cfg.export.edl_frame_rate, 0.04)
+    best: tuple[float, float, int] | None = None
+    with open_video(cfg.paths.source_movie) as cap:
+        for scene_id in scene_ids:
+            scene = next((s for s in scenes if int(s["scene_id"]) == int(scene_id)), None)
+            if scene is None:
+                continue
+            start_s = max(0.0, float(scene["start_s"]) - 0.25)
+            end_s = max(start_s, float(scene["end_s"]) - max(0.04, segment_beat.duration_s) + 0.25)
+            t = start_s
+            while t <= end_s:
+                score = _content_alignment_score(cap, t, templates, cfg)
+                if best is None or score > best[0]:
+                    best = (score, t, int(scene_id))
+                t = round(t + step_s, 6)
+
+    if best is None or best[0] < min_score:
+        return None
+
+    score, in_point_s, scene_id = best
+    duration_s = max(0.0, min(segment_beat.duration_s, segment_beat.end_s - segment_beat.start_s))
+    return MatchSegment(
+        trailer_offset_s=segment_offset_s,
+        duration_s=duration_s,
+        scene_id=scene_id,
+        in_point_s=in_point_s,
+        out_point_s=in_point_s + duration_s,
+        match_score=score,
+        is_confirmed=score >= cfg.cv.deep_scan.match_threshold,
+    )
+
+
 def cmd_match(args: argparse.Namespace, cfg) -> list:
     from src.pipeline.matcher import run_matching
     from dataclasses import replace
@@ -562,6 +736,7 @@ def cmd_match(args: argparse.Namespace, cfg) -> list:
         force_reindex=args.force_reindex,
         seed_in_points=seed_in_points,
     )
+    results = _match_unmatched_visual_segments(results, beats, cached, cfg)
     results = _attach_visual_segments(results, beats, cfg)
 
     # A targeted one-beat match should improve the cache without deleting

setup_venv.ps1

@@ -44,7 +44,25 @@ if ($pythonVersion -notmatch "3\.(1[1-9]|[2-9]\d)") {
 
 # ---- 2. Create venv ---------------------------------------------------------
 if (Test-Path $VENV_DIR) {
-    Write-Host "Virtual environment already exists at '$VENV_DIR'. Skipping creation." -ForegroundColor Yellow
+    $existingVenvPython = Join-Path $VENV_DIR "Scripts\python.exe"
+    $venvOk = $false
+    if (Test-Path $existingVenvPython) {
+        try {
+            $existingVersion = & $existingVenvPython --version 2>&1
+            $venvOk = $LASTEXITCODE -eq 0 -and $existingVersion -match "3\.(1[1-9]|[2-9]\d)"
+        } catch {
+            $venvOk = $false
+        }
+    }
+
+    if ($venvOk) {
+        Write-Host "Virtual environment already exists at '$VENV_DIR'. Skipping creation." -ForegroundColor Yellow
+    } else {
+        Write-Host "Existing virtual environment is not usable. Recreating '$VENV_DIR' ..." -ForegroundColor Yellow
+        Remove-Item -LiteralPath $VENV_DIR -Recurse -Force
+        & $PROJECT_PYTHON -m venv $VENV_DIR
+        Write-Host "Done." -ForegroundColor Green
+    }
 } else {
     Write-Host "Creating virtual environment in '$VENV_DIR' ..." -ForegroundColor Green
     & $PROJECT_PYTHON -m venv $VENV_DIR