macromate-exact-fix

Nutrition prediction returns 0 kcal when the extractor fails to read nested nutrition_data fields correctly. Fixing the Django extractor to normalize calories, protein, and carbohydrates ensures accurate values instead of defaulting to zero.

Loading EfficientNetV2-M in PyTorch for nutrition prediction requires configuring the model loader for 351 classes and 480x480 input dimensions. This stops predictions from incorrectly falling back to CLIP.

Fixing the CLIP fallback issue involves switching the model loader to the correct EfficientNetV2-M architecture with 351 classes. This ensures the primary model handles predictions directly instead of delegating to CLIP.

Yes, EfficientNetV2-M requires a label_nutrition_mapping.json file containing 351 entries. Startup validation verifies these entries match the model classes and logs known foods to keep results reliable.

A Django nutrition prediction pipeline requires PyTorch and torchvision to load EfficientNetV2-M and handle 480x480 input preprocessing. A 351-entry nutrition mapping JSON is also required for startup validation.

Validating a nutrition map at Django startup involves verifying the entries in label_nutrition_mapping.json against the 351 expected classes. This process logs known foods to ensure the pipeline remains reliable.