Plate

About The Project

Photograph a meal, correct the portion, log it.

Plate is a calorie and macro log whose primary input is a photo. The multimodal vision model identifies the food on the plate; the person who ate it corrects the weight. That division of labour is not a convenience — it is the core finding the entire design rests on.

Benchmarked across 145 plates from the Nutrition5k dataset with weighed ground truth, visual human correction was proven to recover roughly half of the accuracy gain of a physical scale, creating an ideal balance between low daily friction and nutritional precision.

Features

• 📸 Vision-First Food Logging: Snap a photo and let the vision model identify foods, decompose composite meals, and estimate baseline nutrition instantly.
• ⚖️ Empirical Portion Calibration: Transparent three-tier error bands based on whether weight was scale-measured, user-adjusted by eye, or model-estimated.
• 🔄 "Not what you're eating?" Re-Read: State what a dish actually is to re-analyze the in-memory photo from scratch without taking a second picture.
• 🏷️ Stated Nutrition Validation: Direct manual entry for printed panels or restaurant menus with automated sanity checks against physically impossible caloric densities.
• 📊 Energy & Weight Trends: Dense, honest daily time-series visualization tracking caloric intake stacked by macronutrient energy alongside least-squares expenditure trends.
• 📦 Complete Data Ownership & Zero Lock-in: Gated-free exports yielding structured JSON, spreadsheet-ready CSV, and a self-contained ZIP archive containing every captured meal photo.
• 🦖 Bitey Nutrition Companion: An encouraging daily companion that celebrates hitting your protein and caloric goals with actionable, personalized feedback.
• 🔒 Private & Account Budgeted: Strictly zero-telemetry architecture with built-in per-account daily quota guards to ensure sustainable operational costs.

Gallery