FitFood is a multi-label image classification system designed to identify over 40 types of Indonesian traditional foods from photographs, built as part of a health-related mobile application.
The model leverages Transfer Learning with InceptionV3 as the base architecture, with fine-tuning of 60+ top layers to adapt the pre-trained features specifically for Indonesian food recognition.
For mobile deployment, the trained model is converted to TensorFlow Lite (.tflite) format, enabling real-time on-device inference on Android smartphones without requiring network connectivity for predictions.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ DATASET │────▶│ InceptionV3 │────▶│ FINE-TUNE │
│ 40+ Foods │ │ Base Model │ │ 60+ layers │
└──────────────┘ └──────────────┘ └──────┬───────┘
│
▼
┌──────────────┐
│ CONVERT │
│ → .tflite │
└──────┬───────┘
│
▼
┌──────────────┐
│ ANDROID APP │
│ On-Device │
└──────────────┘Transfer Learning → Fine-Tuning → TFLite Conversion → Mobile Deployment