scorevision: push artifact

miner.py

@@ -3,15 +3,16 @@
 Base weights: plate_v3 (YOLO26s fine-tuned on Roboflow-filtered + 10x live pseudo-GT,
 resumed from plate_v2). fp16 end2end ONNX, static 1x3x1280x1280, ~19.4 MB.
 
-Inference pipeline (tuned per bench_v2.py + live observation on first 5 shards):
-  - Single full-image pass with soft-NMS + hflip TTA
+Inference pipeline (recall-biased, latency-optimized):
+  - Single full-image forward pass + soft-NMS (NO hflip TTA — drop saves ~1s
+    to reduce p95 variance; TEE chute sees 10s gate, we need headroom)
   - Recall-biased preset: conf=0.22, iou=0.41, sigma=0.685, max_det=22
-  - No tile fallback (v3's recall is already high without tiles)
+  - No tile fallback
 
-Bench on 184-shard live pseudo-GT pool (/mnt/shadeform-data/plate_research/live_gt/):
-  gated=0.440  mAP=0.978 (highest)  fp/img=0.38  ms_p95=157
-Switched from c30 after live shard be77593656fa: we scored 0.168 (mAP 0.500)
-while competitors hit 0.318 (mAP 0.750) — missed borderline-conf plates.
+Bench (c=0.22 without TTA, estimated): gated ≈ 0.436, mAP ≈ 0.975
+Trade +0.005 gated for ~1s faster median / safer p95 vs 10s validator gate.
+Prior LAT events (p95 11.4s and 99s cold-start) showed tail events dominate
+when median is 2-3s; cutting 1 forward pass pulls median to ~1.5s.
 Compared to:
   plate_v2 best:     gated=0.424
   hermestech best:   gated=0.422
@@ -127,7 +128,7 @@ class Miner:
         self.iou_thres = 0.41
         self.sigma = 0.685
         self.max_det = 22
-        self.use_tta = True
+        self.use_tta = False  # disabled: single forward pass, half the latency
 
         print(f"ONNX model loaded from: {model_path}")
         print(f"ONNX providers: {self.session.get_providers()}")