Agricultural robots are an important part of the development of agricultural modernization, and computer vision technology effectively promotes their application in the field of agriculture by perceiving and analyzing crops and the environment. However, because of the complexity and diversity of agricultural scenes, the detailed and annotated large-scale image datasets required by advanced computer vision methods are scarce in the field of agriculture. This lack of datasets is the main challenge in the development of computer vision technology in the field. To solve this problem, this paper presents a large-scale tomato image dataset that can be used for semantic image segmentation, instance segmentation, target detection, and other tasks. The dataset consists of synthetic and real images. The synthetic images include 3250 synthetic tomato images and the corresponding pixel-level semantic segmentation label images; the real images consist of 750 monocular images and 400 binocular images taken by RGB cameras, some of which have detailed manual labels for instance segmentation and target detection. This research aims to enrich many aspects of the dataset, including its capacity, the dimensionality of the annotation information, and the complexity of the scene, and to provide data support for solving future problems in the field of agriculture using computer vision technology.