Search papers, labs, and topics across Lattice.
This paper introduces two novel datasets, BUTom21 and BUTom-ST21, designed for the visual sensing of tomato plants, featuring both still images with manual annotations and video data with semi-automated pseudo-labels. The datasets provide pixel-level labels indicating the ripeness of the fruit, facilitating advanced research in detecting, segmenting, and tracking tomato plants in commercial-like environments. By offering a challenging set of real-world imagery, this work aims to enhance field-based phenotyping methods crucial for horticultural advancements.
Real-world datasets reveal that tracking tomato ripeness can significantly enhance agricultural automation and phenotyping techniques.
In this manuscript we release two datasets for visual sensing of tomato plants grown in commercial-like settings and acquired using a robot. The first is BUTom21 which consists of still images and manual annotations. The second is BUTom-ST21 which consists of video-based data and semi-automated annotations through AI-based methods, referred to as pseudo-labels. In both cases, we provide pixel-level labels for the ripeness of the fruit. The aim is to provide the research community a challenging set of real-world imagery to explore methods to sense and estimate the state of tomato plants and their fruit, which is an important horticultural crop. Importantly, the spatial-temporal dataset provides individual fruit count and ripeness information enabling researchers to push the boundaries of field-based phenotyping.
