Related Experiment Video
Updated: May 2, 2026
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Evaluating food portion estimation accuracy with multi-angle photographs
1Department of Food and Nutrition, Kongju National University, Yesan 32439, Korea.
Background/Objectives:
This study aimed to evaluate the validity of estimating food quantities using photographs taken at different angles to increase the accuracy of dietary intake surveys.
Subjects/Methods:
Eighty-two adults (41 males and 41 females), ranging in age from their 20s to 50s, participated in the study. The participants observed 6 types of food-cooked rice, soup, grilled fish, vegetables, kimchi, and beverages-arranged to simulate an actual meal. After a 3-min observation, they were asked to move to another room and select a photograph that they believed matched the observed food amount. Photographs of each food were taken from 3 different angles (0°, 45°, 70° for solid foods; 45°, 60°, 70° for beverages). The accuracy, underestimation, and overestimation rates were calculated for each type of food and angle.
Results:
Cooked rice had the highest accuracy at 45° (74.4%) (P < 0.001), which improved to 85.4% when multiple angles were combined. Soup showed lower accuracy across all angles and had higher overestimation rates. The angles for the grilled fish did not show significant differences, but the accuracy slightly improved when the angles were combined. For vegetables, the accuracy increased to 53.7% when the angles were combined (P < 0.05). Kimchi showed the highest accuracy at 45° (52.4%), and beverages showed the highest accuracy at 70° (73.2%).
Conclusion:
The accuracy of food quantity estimation varies depending on the type of food and the shooting angle. For solid foods, 45° provided the best accuracy, whereas 70° was most accurate for beverages. Combining different angles improved the estimation accuracy for most food types.
