Baseline CNN RGB + Depth Carbohydrate Regression

This model predicts dish-level carbohydrate content from overhead RGB images and overhead depth images from Nutrition5K.

Dataset

  • Based on a filtered official split universe
  • Train: 2421 dishes
  • Validation: 334 dishes
  • Test: 507 dishes

Architecture

  • Dual-input CNN
  • One RGB branch
  • One depth branch
  • Feature fusion head
  • Dense regression output

Target

  • total_carb
Downloads last month
4
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support