Upload a photo of waste and the model will identify its category across 18 waste types. Powered by EfficientNet-B0 (4M params · 95.1% accuracy) trained on RealWaste + custom dataset.