Food Calorie Recognition API

A FastAPI backend that accepts food images from a Flutter app, classifies them using a TFLite (or Keras MobileNetV2) model, and returns nutrition data from the Edamam Food Database API.

Features

• Image classification — runs a TFLite model if one is provided, otherwise falls back to ImageNet-pretrained MobileNetV2 via Keras
• Nutrition lookup — queries the Edamam Food Database API for calories, protein, fat, carbs, and fiber per 100 g
• Flutter-ready response format — structured JSON with a success flag and a data envelope
• Graceful degradation — Edamam rate limits (429) and outages (503) return a response without nutrition rather than an error

Requirements

• Python 3.11+
• Dependencies listed in requirements.txt

Setup

# 1. Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Copy the environment template and fill in your values
cp .env.example .env

Environment variables

Variable	Default	Description
APP_NAME	IN2 Food Recognition API	API title shown in docs
DEBUG	false	Enable debug mode
MODEL_PATH	(empty)	Path to .tflite model file; leave empty to use Keras fallback
LABELS_PATH	(empty)	Path to labels text file (one label per line)
TOP_K	5	Number of top predictions to return
INPUT_SIZE	224	Input image size (width × height) in pixels
EDAMAM_APP_ID	(empty)	Edamam API app ID — get one at developer.edamam.com
EDAMAM_APP_KEY	(empty)	Edamam API app key

Model files

Place your TFLite model and labels file in the models/ directory and set MODEL_PATH / LABELS_PATH accordingly. If no TFLite model is found, the server automatically downloads and uses the ImageNet MobileNetV2 weights from Keras on first startup.

A helper script is available to download a pre-trained model:

python scripts/download_model.py

Running the server

uvicorn app.main:app --reload

The API is available at http://localhost:8000. Interactive docs are at http://localhost:8000/docs.

API reference

GET /

Returns a welcome message.

GET /health

Returns server status, active model type, and whether Edamam is configured.

{
  "status": "ok",
  "model": "tflite",
  "edamam_configured": true
}

POST /analyze

Accepts a food image and returns classification predictions with optional nutrition data.

Request — multipart/form-data

Field	Type	Description
file	file	JPEG, PNG, GIF, WebP, or BMP image

Response

{
  "success": true,
  "data": {
    "top_food": "pizza",
    "confidence": 0.8532,
    "predictions": [
      { "label": "pizza", "confidence": 0.8532 },
      { "label": "hamburger", "confidence": 0.0981 }
    ],
    "nutrition": {
      "food_id": "food_a1gb90bazibb6bapkfzjbaxxxx",
      "label": "Pizza",
      "category": "Generic foods",
      "calories_per_100g": 266,
      "protein_per_100g": 11.0,
      "fat_per_100g": 10.4,
      "carbs_per_100g": 33.0,
      "fiber_per_100g": 2.3
    },
    "model_type": "tflite"
  },
  "timestamp": null
}

nutrition is null if Edamam is not configured or the lookup fails.

Running tests

pytest

Tests mock both the classifier and the Edamam service so no model files or API keys are required.

Project structure

app/
  main.py              # FastAPI app, lifespan setup, route handlers
  config.py            # Settings loaded from environment variables
  classifier/
    food_classifier.py # TFLite / Keras MobileNetV2 inference
  services/
    edamam.py          # Edamam Food Database API client
models/
  food_classifier.tflite
  labels.txt
scripts/
  download_model.py    # Helper to download a pre-trained TFLite model
  train_food_model.py  # Training script
tests/
  test_api.py
  test_classifier.py
  test_edamam.py
  test_flutter.py

IN2 Food Recognition API

A FastAPI backend that accepts food images from a Flutter app, classifies them using a TFLite (or Keras MobileNetV2) model, and returns nutrition data from the Edamam Food Database API.

Features

• Image classification — runs a TFLite model if one is provided, otherwise falls back to ImageNet-pretrained MobileNetV2 via Keras
• Nutrition lookup — queries the Edamam Food Database API for calories, protein, fat, carbs, and fiber per 100 g
• Flutter-ready response format — structured JSON with a success flag and a data envelope
• Graceful degradation — Edamam rate limits (429) and outages (503) return a response without nutrition rather than an error

Requirements

• Python 3.11+
• Dependencies listed in requirements.txt

Setup

# 1. Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Copy the environment template and fill in your values
cp .env.example .env

Environment variables

Variable	Default	Description
APP_NAME	IN2 Food Recognition API	API title shown in docs
DEBUG	false	Enable debug mode
MODEL_PATH	(empty)	Path to .tflite model file; leave empty to use Keras fallback
LABELS_PATH	(empty)	Path to labels text file (one label per line)
TOP_K	5	Number of top predictions to return
INPUT_SIZE	224	Input image size (width × height) in pixels
EDAMAM_APP_ID	(empty)	Edamam API app ID — get one at developer.edamam.com
EDAMAM_APP_KEY	(empty)	Edamam API app key

Model files

Place your TFLite model and labels file in the models/ directory and set MODEL_PATH / LABELS_PATH accordingly. If no TFLite model is found, the server automatically downloads and uses the ImageNet MobileNetV2 weights from Keras on first startup.

A helper script is available to download a pre-trained model:

python scripts/download_model.py

Running the server

uvicorn app.main:app --reload

The API is available at http://localhost:8000. Interactive docs are at http://localhost:8000/docs.

API reference

GET /

Returns a welcome message.

GET /health

Returns server status, active model type, and whether Edamam is configured.

{
  "status": "ok",
  "model": "tflite",
  "edamam_configured": true
}

POST /analyze

Accepts a food image and returns classification predictions with optional nutrition data.

Request — multipart/form-data

Field	Type	Description
file	file	JPEG, PNG, GIF, WebP, or BMP image

Response

{
  "success": true,
  "data": {
    "top_food": "pizza",
    "confidence": 0.8532,
    "predictions": [
      { "label": "pizza", "confidence": 0.8532 },
      { "label": "hamburger", "confidence": 0.0981 }
    ],
    "nutrition": {
      "food_id": "food_a1gb90bazibb6bapkfzjbaxxxx",
      "label": "Pizza",
      "category": "Generic foods",
      "calories_per_100g": 266,
      "protein_per_100g": 11.0,
      "fat_per_100g": 10.4,
      "carbs_per_100g": 33.0,
      "fiber_per_100g": 2.3
    },
    "model_type": "tflite"
  },
  "timestamp": null
}

nutrition is null if Edamam is not configured or the lookup fails.

Running tests

pytest

Tests mock both the classifier and the Edamam service so no model files or API keys are required.

Project structure

app/
  main.py              # FastAPI app, lifespan setup, route handlers
  config.py            # Settings loaded from environment variables
  classifier/
    food_classifier.py # TFLite / Keras MobileNetV2 inference
  services/
    edamam.py          # Edamam Food Database API client
models/
  food_classifier.tflite
  labels.txt
scripts/
  download_model.py    # Helper to download a pre-trained TFLite model
  train_food_model.py  # Training script
tests/
  test_api.py
  test_classifier.py
  test_edamam.py
  test_flutter.py