petrol-fastwin-v6 (INT8 + WBF + agreement gate)
Same approach as v5, with INT8 dynamic quantization to fit the
30 MB repo size limit for manak0/Detect-petrol-station-1-0.
weights/havengo.onnxโ INT8 quantized (22.7 MB โ 11.6 MB)weights/superbit.onnxโ INT8 quantized (19.4 MB โ 10.1 MB)- Total: 21.8 MB (limit 30 MB)
WBF ensemble + cluster-agreement gate (both models must propose a box within WBF IoU 0.55) โ the +12.2% composite recipe from v5.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support