Semi-outdoor object detection scenarios dataset

This benchmark dataset contains over 2,900 semi-outdoor images collected from real-world intelligent waste collection systems, supporting two application scenarios: person detection and garbage detection. The Person Detection dataset comprises 1,616 images covering 17 categories, capturing human interactions with disposed items around waste collection units. The Garbage Detection dataset comprises 1,328 images covering 16 categories, focusing on waste classification inside waste containers. All images exhibit variations in illumination conditions (sunny, cloudy, and dusk), background complexity (open areas, vegetation regions, and building surfaces), and object scales ranging from large foreground instances to distant targets occupying only a few pixels. The two datasets offer complementary perspectives: the Person Detection dataset emphasizes macro‑level external scenes and human‑object interactions, while the Garbage Detection dataset provides a micro‑level view inside containers for fine‑grained waste recognition. Each dataset is further divided into Sample‑A and Sample‑B subsets, representing limited‑data versus larger‑data configurations without altering the inherent visual characteristics.