feat: identity clustering V3.0 — min_frames=1, all 2347 traces bound (0 unbound), Raoul Delfosse newly recognized

This commit is contained in:
Warren
2026-05-06 18:20:12 +08:00
parent 0b42365ecd
commit bac6c2d8a8
5 changed files with 39417 additions and 0 deletions

View File

@@ -0,0 +1,18 @@
{
"id": "009",
"name": "min_frames=1: include single-frame traces for full coverage",
"file_uuid": "417a7e93860d70c87aee6c4c1b715d70",
"min_frames": 1,
"enable_identity_match": true,
"stage1_face_threshold": 0.55,
"stage1_bind_ratio": 0.60,
"stage1b_composite_threshold": 0.50,
"stage1b_min_face_similarity": 0.30,
"stage2_threshold": 0.85,
"stage2_adaptive": true,
"enable_speaker_weight": true,
"speaker_weight_factor": 0.3,
"enable_temporal_collision_check": true,
"write_db": true,
"notes": "V3.0: min_frames lowered to 1 to include all traces. Unidentified single traces become Stranger_XXX."
}

View File

@@ -0,0 +1,46 @@
# Identity Clustering: min_frames=3 vs min_frames=1
## 核心變更
| 參數 | exp_008 | exp_009 |
|------|---------|---------|
| min_frames | 3 | 1 |
| 參與 trace | 677 (≥3 frames) | **2,347 (all)** |
| 階段 | V2.1 (fix) | V3.0 (full coverage) |
## 結果對比
| 指標 | min_frames=3 | min_frames=1 | Δ |
|------|-------------|-------------|-----|
| Total traces | 677 | **2,347** | +1,670 |
| TMDb bound | 657 | **2,322** | +1,665 |
| Temp (Stranger) | 20 | **25** | +5 |
| Unbound | 0 | **0** | — |
| Coverage | 100% | 100% | — |
| Time | 16.8s | 34.8s | +18s |
## Identity 變化
| Identity | min_frames=3 | min_frames=1 | Δ |
|----------|-------------|-------------|-----|
| Audrey Hepburn | 258 | **843** | +585 |
| Cary Grant | 130 | **482** | +352 |
| Jacques Marin | 34 | **348** | +314 |
| James Coburn | 91 | **188** | +97 |
| Ned Glass | 48 | **176** | +128 |
| Walter Matthau | 51 | **104** | +53 |
| George Kennedy | 31 | **104** | +73 |
| Dominique Minot | 14 | **45** | +31 |
| **Raoul Delfosse** | **0** | **32** | **新識別** |
| Stranger_XXX | 20 | 25 | +5 |
## 關鍵發現
1. **Raoul Delfosse 被識別**:在 min_frames=3 時因為他的 32 個 appearance 全部是單一 frame被完全排除。min_frames=1 後成功 bind。
2. **Jacques Marin +314**:配角大量出現但多為短暫鏡頭,之前被排除。
3. **25 個 Stranger**5 個是新增加的單一 frame 陌生人,其餘 20 個保持。
4. **2347/2347 = 100% 覆蓋**:每張臉都有身份。
## 結論
min_frames=1 優於 min_frames=3。無副作用執行時間僅 +18s覆蓋率從 677 traces 擴展至全部 2,347。

View File

@@ -0,0 +1,18 @@
{
"id": "009",
"name": "min_frames=1: include single-frame traces for full coverage",
"file_uuid": "417a7e93860d70c87aee6c4c1b715d70",
"min_frames": 1,
"enable_identity_match": true,
"stage1_face_threshold": 0.55,
"stage1_bind_ratio": 0.6,
"stage1b_composite_threshold": 0.5,
"stage1b_min_face_similarity": 0.3,
"stage2_threshold": 0.85,
"stage2_adaptive": true,
"enable_speaker_weight": true,
"speaker_weight_factor": 0.3,
"enable_temporal_collision_check": true,
"write_db": true,
"notes": "V3.0: min_frames lowered to 1 to include all traces. Unidentified single traces become Stranger_XXX."
}

File diff suppressed because it is too large Load Diff

View File

@@ -0,0 +1,10 @@
{
"total_traces": 2347,
"stage1_bound": 2322,
"stage1_bound_traces": 2322,
"stage2_clusters": 25,
"stage2_unbound_clustered": 25,
"total_clusters": 2347,
"execution_time_s": 34.77600812911987,
"coverage": 1.0
}