All projects

AI Food Recognition & Nutrition Estimation

Photograph a meal and the system detects each food, matches it to Chinese dishes and logs its nutrition. Any prediction the model is unsure of waits for the user’s confirmation.

Jun 2026Independent build

Demo recordings

Desktop

Mobile

Recognition pipeline

Photo uploadDesktop or phone camera
Region detectionYOLOv8 FoodSeg103
Dish matchingChinese-CLIP candidates
NutritionLocal database × adjustable portion
Confidence gateUnsure items go to review
Meal recordsHistory, daily goals, weekly charts

Human-in-the-loop review

Uncertain predictions never enter the log automatically.

Detected food regionsYOLOv8 boxes + Chinese-CLIP candidates
Portion sizeUser-adjustable grams
Confidence checkReliable enough to record?
High confidenceRecorded with calories and macros
Low confidenceReview queue until the user confirms
Not recognisedManual-entry fallback

Selected screens

Multi-region detection
Multi-region detection
Low-confidence review queue
Low-confidence review queue
Mixed dish
Mixed dish
Single-food recognition
Single-food recognition
Daily tracker and history
Daily tracker and history
Weekly intake chart
Weekly intake chart

Known limitations

Overlapping ingredientsMixed dishes blur region boundaries.
Visual similarityLook-alike dishes get mismatched.
No depth or scaleA single 2D photo cannot reveal portion size.
EV Population AnalysisDiabetes Readmission