petrol-fastwin-v5 (WBF + agreement gate)
Iteration of v3. Same two-detector WBF ensemble:
weights/havengo.onnxโ m50 0.901 / fp 0.884 (high recall)weights/superbit.onnxโ m50 0.892 / fp 0.944 (high precision)
v5 change: cluster_size โฅ 2 โ require both models to propose a box (within WBF IoU 0.55) for it to survive. v3 hurt the fp pillar by promoting singleton clusters (only one model agreed) into the output; this drops them structurally.
Per-model weights 0.45 / 0.55 favor SuperBit slightly.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support