FoodSeg103

In this work, we build a new food image dataset FoodSeg103 (and its extension FoodSeg154) containing 9,490 images. We annotate these images with 154 ingredient classes and result in an average of 6 ingredient labels and pixel-wise masks per image. In addition, we propose a multi-modality pre-training approach called ReLeM that explicitly equips the model with rich and semantic food knowledge. In experiments, we use three popular semantic segmentation methods (i.e., Dilated Convolution based, Feature Pyramid based, and Vision Transformer based) as baselines, and evaluate them as well as ReLeM on our new datasets. We believe that the FoodSeg103 (and its extension FoodSeg154) and the pre-trained models using ReLeM can serve as a benchmark to facilitate future works in fine-grained food image understanding.

在本项研究中，我们构建了一个新的食品图像数据集FoodSeg103（及其扩展FoodSeg154），包含9,490张图像。我们对这些图像进行了154种成分类别的标注，每个图像平均包含6个成分标签和像素级掩码。此外，我们提出了一种名为ReLeM的多模态预训练方法，该方法明确地为模型配备了丰富且语义化的食品知识。在实验中，我们以三种流行的语义分割方法（即基于扩张卷积、基于特征金字塔和基于视觉Transformer的方法）作为基准，并对它们以及ReLeM在我们的新数据集上的表现进行了评估。我们相信，FoodSeg103（及其扩展FoodSeg154）以及使用ReLeM预训练的模型可以作为基准，以促进未来对细粒度食品图像理解的研究。