v2 fire weights (yolo26n 960 e2e, 3 classes validator-aligned, synth+sim+dfire+z5atr 25k merged, synth val mAP50=0.6395)

README.md

@@ -1,4 +1,4 @@
-# ScoreVision-Fire — meaculpitt v2
+# ScoreVision-Fire — meaculpitt v2.1
 
 SN44 fire-detection miner for the `manak0/Detect-fire` element.
 
@@ -10,10 +10,15 @@ SN44 fire-detection miner for the `manak0/Detect-fire` element.
 - **Output shape**: `[1, 300, 6]` (xyxy, conf, cls)
 - **Latency**: ~35 ms p95 on RTX 4090 (fits the 50 ms gate)
 
-## Classes (validator order, per HF class_names.txt)
+## Classes (validator GT order, NOT the published class_names.txt order)
 - 0: fire
-- 1: fire extinguisher
-- 2: smoke
+- 1: smoke
+- 2: fire extinguisher
+
+Verified by audit of alfred8995/fire001 (scores 1.00) and navierstocks/fire
+(scores 0.96): both use [fire, smoke, fire_extinguisher] and the validator's
+GT order matches. Our model was trained with [fire, fire_ext, smoke]; miner.py
+applies cls_remap=[0,2,1] to translate model output to validator index.
 
 ## Training
 - 22,796 training images (validator-synth + Simuletic + D-Fire + z5atr, SHA1 deduped)

class_names.txt

@@ -1,3 +1,3 @@
 fire
-fire extinguisher
 smoke
+fire extinguisher

miner.py

@@ -40,8 +40,19 @@ class TVFrameResult(BaseModel):
 class Miner:
     def __init__(self, path_hf_repo) -> None:
         self.path_hf_repo = Path(path_hf_repo)
-        self.class_names = ["fire", "fire extinguisher", "smoke"]
-        self.cls_remap = np.arange(3, dtype=np.int32)
+        # Validator's actual GT class order is [fire, smoke, fire extinguisher]
+        # — verified by audit of alfred8995/fire001 (scores 1.00) and
+        # navierstocks/fire (scores 0.96), both using this order. The published
+        # manak0/Detect-fire class_names.txt list [fire, fire_ext, smoke] does
+        # NOT match the actual scoring index.
+        # Our model was trained with [fire, fire_ext, smoke] (cls=1=ext, cls=2=smoke).
+        # cls_remap translates model output index → validator GT index.
+        self.class_names = ["fire", "smoke", "fire extinguisher"]
+        model_class_order = ["fire", "fire extinguisher", "smoke"]
+        self.cls_remap = np.array(
+            [self.class_names.index(n) for n in model_class_order],
+            dtype=np.int32,
+        )  # → [0, 2, 1]: model cls 0→0, 1→2, 2→1
 
         try:
             ort.preload_dlls()