OAM-3D: A Multimodal 3D Grape Detection Dataset and Code for Occlusion-Aware Orchard Perception

This record provides the processed multimodal 3D grape dataset and source code used in the manuscript “Occlusion-Aware Dual-Expert Fusion for Multimodal 3D Grape Detection in Orchard Environments”. The dataset contains synchronized RGB images, point clouds, 3D annotations, scene-level splits, and occlusion-related attributes for orchard grape detection. The accompanying code implements the OAM-3D framework, including occlusion auxiliary supervision, BackboneFiLM, dual-expert cross-attention fusion, and a size-aware bounding-box coder.