DWaste: Greener AI for Waste Sorting using Mobile and Edge Devices

The rise of the convenience packaging has led to generation of enormous waste. Sorting these waste is crucial in waste management. To aid the  sorting process we provide an AI powered platform call DWaste that runs on smartphones and edge devices, even without internet access. We trained image classification (EfficientNetV2S/M, ResNet50/101) and light weighted object detection (YOLOv8n, YOLOv11n)  models using a subset of our own Waste dataset and  annotated it using the custom tool Annotated Lab. Among these models, EfficientNetV2S  achieved 96% accuracy but took longer inference time (0.22s) and higher carbon emission, whereas the lightweight object detection delivered 76–77% mAP, with low emissions, ultra-fast inference (~0.03s) with  smaller model sizes(<7MB), making them ideal for real-time use on low-power devices. Quantization further reduced model size and VRAM usage by nearly half. Our experiment show how greener AI model support real-time waste sorting to manage waste.