Climate ToolsetHow Food Carbon Photo Estimator works
Press Load model once: your browser downloads EfficientNet-Lite0 image classifier (MediaPipe, int8) (17 MB) from this site and keeps it in its cache. Then add your file and press Estimate food category. The model runs on this device, so nothing you enter is sent to a server. Until the model is loaded, or if your device cannot run it, the tool falls back to a simple keyword check and labels the result as the fallback.
EfficientNet-Lite0 image classifier (MediaPipe, int8) names the main subject of a photo from the 1,000 ImageNet categories, such as "water bottle", "carton" or "laptop", with a confidence for its top five guesses.
The tool then maps those labels onto its own groups (listed below): the first of the top guesses, in confidence order and above 5%, that belongs to a group decides the result. If none does, it says it is not sure rather than guess.
Groups
- Higher-carbon: red meat - Beef, lamb and processed red meat are usually the highest-emission foods per serving. The tier is a rough food category, not a measured footprint: farming method, portion, transport and waste all change the real number.
- Medium: dairy, poultry, fish - Cheese, eggs, poultry, seafood and chocolate usually sit between red meat and plant foods. The tier is a rough food category, not a measured footprint: farming method, portion, transport and waste all change the real number.
- Lower-carbon: plants - Fruit, vegetables, grains and bread are usually the lowest-emission foods. The tier is a rough food category, not a measured footprint: farming method, portion, transport and waste all change the real number.
