M.S. AAI Capstone Chronicles 2024

milestones. More diverse food dishes from different cuisines should be added to boost accuracy for rare scenarios.

Figure 3.2: A pre-trained YOLOv8 backbone was used for feature extraction from the food image. A Dense network was then used as a regression head to generate the predictions.

Figure 3.3: Training and validation results of the YOLOv8

22

256