arXiv:2408.05445v1 Announce Type: new
Abstract: Current research in food analysis primarily concentrates on tasks such as food recognition, recipe retrieval and nutrition estimation from a single image. Nevertheless, there is a significant gap in exploring the impact of food intake on physiological indicators (e.g., weight) over time. This paper addresses this gap by introducing the DietDiary dataset, which encompasses daily dietary diaries and corresponding weight measurements of real users. Furthermore, we propose a novel task of weight prediction with a dietary diary that aims to leverage historical food intake and weight to predict future weights. To tackle this task, we propose a model-agnostic time series forecasting framework. Specifically, we introduce a Unified Meal Representation Learning (UMRL) module to extract representations for each meal. Additionally, we design a diet-aware loss function to associate food intake with weight variations. By conducting experiments on the DietDiary dataset with two state-of-the-art time series forecasting models, NLinear and iTransformer, we demonstrate that our proposed framework achieves superior performance compared to the original models. We make our dataset, code, and models publicly available at: https://yxg1005.github.io/weight-prediction/.

Exploring the Relationship Between Food Intake and Weight with the DietDiary Dataset

Food analysis has been an active area of research in recent years, with topics such as food recognition, recipe retrieval, and nutrition estimation from images gaining a lot of attention. However, there has been a lack of focus on understanding how food intake impacts physiological indicators like weight over time. In this paper, the authors bridge this gap by introducing the DietDiary dataset, which consists of daily dietary diaries and corresponding weight measurements from real users.

One of the key contributions of this work is the introduction of a novel task called weight prediction with a dietary diary. The goal of this task is to leverage historical food intake and weight data to predict future weights. This is a challenging problem as it requires modeling the complex relationships between food intake and weight variations over time.

To tackle this task, the authors propose a model-agnostic time series forecasting framework. This framework consists of two main components: the Unified Meal Representation Learning (UMRL) module and the diet-aware loss function. The UMRL module extracts representations for each meal, capturing important features that may influence weight variations. The diet-aware loss function associates food intake with weight variations, enabling the model to learn the relationship between the two.

The authors evaluate their proposed framework on the DietDiary dataset using two state-of-the-art time series forecasting models, NLinear and iTransformer. The experimental results show that the proposed framework outperforms the original models, indicating the effectiveness of the UMRL module and the diet-aware loss function in capturing the dynamics of food intake and weight.

This research is highly interdisciplinary, combining concepts from various fields such as multimedia information systems, animations, artificial reality, augmented reality, and virtual realities. The integration of these fields allows for a deeper understanding of the relationship between food intake and weight variations over time.

Overall, this paper makes significant contributions to the field of food analysis by introducing the DietDiary dataset and proposing a novel task and framework for weight prediction with dietary diaries. The availability of the dataset, code, and models will undoubtedly benefit further research in this area. Moving forward, it would be interesting to see how this framework can be extended to other physiological indicators and how it performs on larger and more diverse datasets.

References:

Read the original article