MEPWaste:A multimodal express packaging waste image dataset

Dataset Description We construct a multimodal dataset of express packaging waste in unstructured conveyor-belt sorting environments, named MEPWaste (Multimodal Express Packaging Waste), which includes both RGB images and depth images, with a total of 4,528 synchronized RGB–depth paired samples (9,056 images). Data acquisition is conducted using an industrial high-resolution RGB camera and a depth camera mounted above a conveyor belt, triggered synchronously to ensure spatial–temporal alignment consistency between modalities. The dataset covers ten representative categories of express packaging waste, including outer packaging (PaperBox, FileSealing, PlasticBag, WovenBag, FoamBox) and inner packaging (GasColumnBag, AirBag, EPEFoam, BubbleFilm, IceBag). Samples are collected under realistic sorting conditions characterized by material heterogeneity, deformable packaging, random stacking, and background interference, reflecting engineering-relevant operational disturbances. To support occlusion-centric benchmarking, MEPWaste intentionally includes images under varying levels of stacking-induced occlusion (light occlusion, moderate occlusion, severe occlusion). Occlusion levels are curated based on the degree of visible-evidence loss caused by object overlap and partial coverage in conveyor-belt scenes. The dataset provides instance-level bounding-box annotations for all categories, enabling reproducible evaluation of multimodal detection methods under dense stacking and heavy occlusion. As the related paper is currently under review, the dataset will remain temporarily confidential, and only partial data are disclosed here. The complete dataset and full benchmark protocol will be released once the article is accepted. Usage Policy If you plan to use our data in scientific research, we strongly recommend contacting us in advance to seek feedback on experimental design and benchmarking settings. Please consider acknowledging our contributions and citing the associated paper/dataset description. For substantial use of the dataset (e.g., building new benchmarks, releasing derivative annotations, or extending the dataset taxonomy), we encourage discussing potential collaboration and co-authorship in advance.