feat(trip-cluster): coverage detection — completeness flags + outliers

2026-06-27 17:19:06 +02:00
parent e01b30a5b5
commit 8d58ec03a9
2 changed files with 83 additions and 0 deletions
@@ -0,0 +1,35 @@
+from photoflow.core.models import ClusterMember
+from app.clustering import _epoch, _median
+
+
+def coverage_members(candidate, assets_by_id, *, outlier_factor: float = 8.0) -> list:
+    member_ids = [i for i in candidate.member_ids if i in assets_by_id]
+    members = sorted((assets_by_id[i] for i in member_ids), key=lambda a: a.taken_at)
+    epochs = [_epoch(a.taken_at) for a in members]
+    gaps = [b - a for a, b in zip(epochs, epochs[1:])]
+    base = _median(gaps) if gaps else 0.0
+
+    out = []
+    for idx, a in enumerate(members):
+        is_outlier = False
+        if candidate.seed_tag and base > 0:
+            left = epochs[idx] - epochs[idx - 1] if idx > 0 else 0
+            right = epochs[idx + 1] - epochs[idx] if idx < len(members) - 1 else 0
+            nearest = min([g for g in (left, right) if g > 0], default=0)
+            if nearest > outlier_factor * base:
+                is_outlier = True
+        out.append(ClusterMember(
+            cluster_id=0, immich_id=a.immich_id,
+            member_confidence=candidate.confidence, is_outlier=is_outlier,
+            included=True, flagged_coverage=False))
+
+    if candidate.seed_tag and candidate.start_at and candidate.end_at:
+        mset = set(member_ids)
+        for a in sorted(assets_by_id.values(), key=lambda x: x.taken_at):
+            if a.immich_id in mset:
+                continue
+            if candidate.start_at <= a.taken_at <= candidate.end_at:
+                out.append(ClusterMember(
+                    cluster_id=0, immich_id=a.immich_id, member_confidence=0.0,
+                    is_outlier=False, included=False, flagged_coverage=True))
+    return out