Towards better small object detection in UAV scenes: Aggregating more object-oriented information

Chenyue Yang, Yichao Cao, Xiaobo Lu

Abstract: Highlights•Low-dimensions are adopted to guide original space to expand the fields-of-view.•Input tokens are represented as wave for dynamic aggregation based on local maximum.•We fuse representations simultaneously by combining CNN and Vision MLP in parallel.•Our proposed algorithm outperforms other popular methods.