SEG-FOOD-Semantic Food Segmentation Through Deep Learning

Semantic segmentation is the topic of interest among deep learning researchers in the recent era. It has many applications in different domains including, food recognition. In the case of food recognition, it removes the non-food background from the food portion. There is no large public food dataset available to train semantic segmentation models. We prepared a dataset named ’SEG-FOOD’[44] containing images of FOOD101, PFID, and Pakistani Food dataset and open-sourced the annotated dataset for future research. We annotated the images using JS Segment annotator. For experimentation, please refer to our paper, and for the starter code, please refer to our Github repository.* Please note that our main contribution is a manual annotation by JS-Segment so that researchers can explore various semantic segmentation based methods based on deep learning. The images of this dataset contain images from Food101, PFID dataset, and our own collected dataset of Pakistani Food.

语义分割（Semantic segmentation）是近年来深度学习研究者关注的热点研究方向。该技术在诸多领域均有广泛应用，食品识别即为其中一类典型场景。在食品识别任务中，语义分割可将图像中的非食品背景剔除，仅保留食品区域。目前尚无大规模公开的食品类数据集可供语义分割模型训练使用。为此，我们构建了名为SEG-FOOD[44]的数据集，该数据集涵盖FOOD101、PFID以及巴基斯坦食品数据集的图像，并将该标注数据集开源，以支持后续相关研究工作。我们采用JS Segment标注工具完成了所有图像的标注。相关实验细节请参阅我们的学术论文，入门示例代码可前往我们的GitHub仓库获取。
* 请注意，本工作的核心贡献在于通过JS-Segment完成的人工标注，以便研究者能够基于深度学习技术探索各类语义分割相关方法。本数据集的图像来源于Food101、PFID数据集以及我们自主采集的巴基斯坦食品数据集。