| Stage | Evidence | Interpretation |
|---|---|---|
| Detector fusion | controlled WBF fallback 0.7587924 hidden | fully reproducible reference |
| Safe crop rescoring | historical protected pipeline 0.7701133 hidden | valid team provenance; byte-exact local reproduction pending |
| Tiny + DINOv3 blend | repository-documented protected approximately-0.789 variant | full-official Y, quantile alignment, grouped80 LambdaRank v2 w=0.7 |
| LambdaRank selection policy | 0.7884920 hidden; +0.0153850 over identical detector/rescorer pipeline | historical protected milestone |
| Owner-verified latest | 0.7910702 hidden · rank 10 / 56 | 29 August team record; exact lineage pending repository synchronization |
| Field | Controlled value |
|---|---|
| Models | Co-DINO 1280 scale-control + MI-DETR 1024 |
| Method | Class-aware Weighted Box Fusion |
| Confidence aggregation | max |
| IoU / skip | 0.45 / 0.001 |
| Model weights | 1.0 / 1.0 |
| Output | stable top-300/image · PURE_INTEGER_XYWH submission |
| Local canonical / AIdea hidden | 0.7468375071854423 / 0.7587924 |
| Status | FULLY REPRODUCIBLE FALLBACK |
Co-DINO1280 + MI-DETR1024
The complete local provenance, prediction artifacts, and equal-weight max-confidence WBF recipe are reconstructed and controlled.
Latest AIdea result
Default branches remain at 31 August (Marine) and 30 August (Overfit). The 0.7910702 score and rank 10 / 56 are the owner-confirmed 29 August record, not a live leaderboard check. No later hidden result, finalist outcome, or active training is established by the reviewed repositories.
| Method | Local mAP50 | AIdea hidden | Decision / interpretation |
|---|---|---|---|
| Co-DINO1280 + MI-DETR1024 · equal-weight WBF | 0.7468375072 | 0.7587924 | FULLY REPRODUCIBLE FALLBACK |
| Co-DINO1280 + RT-DETRv4-X1280 | 0.7414361971 | 0.7415787 | REJECTED AS PRIMARY TWO-MODEL FUSION |
| Co-DINO1280 + MI-DETR1024 + RT-DETRv4-X1280 | 0.7489094849 | — | HISTORICAL LOCAL CANDIDATE · later global-member audit rejected |
| Protected crop-rescorer milestone | — | 0.7701133 | HISTORICAL TEAM PROVENANCE · BYTE-EXACT REPRODUCTION PENDING |
| Owner-verified latest team result | — | 0.7910702 | 29 AUG RECORD · RANK 10 / 56 · 29 AUG · EXACT LINEAGE PENDING |
| Prior 1024 equal-weight WBF | 0.7437401061 | 0.7534597 | HISTORICAL REFERENCE |
| Class-bucket WBF | 0.7445862966 | 0.7534510 | REJECTED · NOT CANONICAL |
| Tiny-area gated RT-DETRv4 | 0.7416704223 | 0.7484452 | REJECTED · HIDDEN REGRESSION |
RT-DETRv4 as third member
The three-model candidate adds approximately 389 unique TP over the two-model base, around 201 under 1% object area and around 28 under 0.1%. It is evidence-supported, not automatically retained.
Quality burden
The later CMY plus RT-DETRv4 and crop-rescorer audit found a small local gain below the integration gate. The global addition was rejected.
Measure replacement value
A model remains only if it helps the reconstructed final system, not merely because it has a good standalone or local-fusion number.
| Experiment | Evaluation scope | Measured result | Decision |
|---|---|---|---|
| MI-DETR1280 E13 · blind replacement | AIdea hidden · same historical pipeline | 0.7689250 vs 0.7701133 · -0.0011883 | REJECTED RECIPE · accepted standalone unchanged |
| MI-DETR1280 E13 · recalibrated / dual-MI grid | val_fit · 39 configurations | 0.771331 vs 0.770556 · +0.000775 | BELOW INTEGRATION GATE · original MI1024 retained |
| CMY + RT-DETRv4 + Tiny/ViT | reported grouped80 experiment | 0.763811 vs 0.762515 · +0.001296 | REJECTED GLOBAL MEMBER · no hidden result |
| Tiny384 letterbox + ViT-S | reported grouped80 experiment | 0.764970 vs 0.762515 · +0.002455 | BELOW INTEGRATION GATE · no hidden result |
| LambdaRank scene context | fixed w=0.7 | 0.775815 vs 0.776482 · -0.000666 | REJECTED · unchanged ranker retained |