Small Object Detection
Overview
Small object detection studies how to locate targets with weak visual detail, such as distant vehicles in drone images, tiny defects, or faint infrared targets. Common approaches preserve high-resolution features, fuse information across scales, and use attention, context, or frequency-based processing to distinguish targets from background; slicing high-resolution images and adding temporal or multimodal cues are also used. Methods are commonly evaluated on aerial and remote-sensing benchmarks such as VisDrone and UAVDT, alongside domain-specific datasets, but performance can change substantially across scenes and target sizes. A central tension is improving recall and localization without excessive computation, while ensuring gains transfer beyond the training dataset and remain reliable when architectural components are combined.
Latest papers
87 papers7 papers in the last four weeks, down 22% on the four weeks before. 0.0% of all new papers.