Year: 2025
Type: article
Abstract: In recent years, the significance of food aesthetics has increasingly been recognized within the food industry and on social media platforms. Accurate evaluation of the aesthetic quality of food images is crucial for enhancing consumer experience and improving marketing outcomes. This study aims to assess the performance of different deep learning models in food aesthetics evaluation to identify the most suitable model for this task. Two mainstream convolutional neural network models, including MobileNetV2 and ResNet50, were employed to score the aesthetics of food images. These models were evaluated based on loss values, mean absolute error (MAE), and aesthetic scores. The experimental results demonstrate that the MobileNetV2 model exhibits higher fitting accuracy and precision in capturing image features and predicting aesthetic scores, with its MAE and loss values both lower than those of the ResNet50 model. Furthermore, the scores generated by MobileNetV2 were closer to the reference results from a benchmark model, further validating its effectiveness in practical applications. Future research could explore the optimization of model structures using Vision Transformers (ViT), which can integrate global and local features for a more comprehensive aesthetic judgment.
DOI: 10.54097/sd34ns56
Author: Dian Zhang
Institution: Beijing Jiaotong University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: diamond
OpenAlex ID: W4407682382