04
Trash Image Classification
Three architectures, one dataset, identical settings — so the comparison actually means something.
- ROLE
- EfficientNet-B0 track & demo
- CONTEXT
- Deep learning coursework · ITC
- STACK
- PyTorch · EfficientNet-B0
- test accuracymy EfficientNet-B0
- 86.5%
- best in groupa teammate's YOLOv8
- 93.2%
- images · six classes
- 2,885
The setup
Coursework split three ways — one model each, trained on the same 2,885-image dataset of six waste classes and compared under identical settings, so the differences between the runs came from the architectures rather than from the training budget.
My track
I took EfficientNet-B0 to 86.5% test accuracy, then built it into a smart-bin demo that classifies whatever you hold up to the webcam in real time. A teammate's YOLOv8 scored highest overall at 93.2%.
- Image classification
- EfficientNet-B0
- Real-time inference
- PyTorch