Rice tillering and plant architecture traits are critical determinants of panicle number, population density, and yield formation. However, field-based measurement of these traits is severely limited by factors such as plant canopy obstruction, uneven lighting, and low efficiency of traditional manual phenotyping. Additionally, existing automated solutions often involve high hardware costs and complex workflows, hindering the development of high-throughput phenotyping methods for these traits. Recently, researchers from the Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, and collaborators developed TillerPET, an AI-based model that enables high-throughput, in-situ phenotyping of rice tiller number and plant compactness using RGB images captured after harvest. The model was trained and validated on a large, multi-year, multi-location rice RGB image dataset, demonstrating consistent performance across diverse environments. TillerPET is built on a point-query Transformer architecture and incorporates a depth-based rice region extraction module. It employs a lightweight feature extraction method that simplifies the encoder structure, significantly reducing computational load while enhancing accuracy. On the tested rice RGB image dataset, TillerPET achieved a coefficient of determination (R²) of 0.941 for tiller counting and an R² of 0.978 for compactness measurement—demonstrating high precision and robustness. By leveraging features extracted by TillerPET, researchers were able to effectively classify different rice genotypes based on their tillering and architectural characteristics. The resulting multi-year and multi-location phenotypic data provide valuable resources for rice plant architecture breeding programs. The study was published in The Crop Journal. The research was supported by the National Natural Science Foundation of China and the Hubei Provincial Natural Science Foundation. This advancement marks a significant step forward in overcoming the challenges of high-throughput phenotyping for key rice traits, paving the way for more efficient and data-driven rice breeding.