Graph fusion network for multi-oriented object detection

In object detection, non-maximum suppression (NMS) methods are extensively adopted to remove horizontal duplicates of detected dense boxes for generating final instances. However, due the degraded quality detection and not explicit exploration context information, existing NMS via simple intersection-over-union (IoU) metrics tend underperform on multi-oriented long-size objects detection. Distinguishing with general duplicate removal, we propose a novel graph fusion network, named GFNet, Our GFNet is extensible adaptively fuse detect more accurate holistic Specifically, first adopt locality-aware clustering algorithm group into different clusters. We will construct an instance sub-graph belonging one cluster. Then, graph-based network Graph Convolutional Network (GCN) learn reason boxes. Extensive experiments both public available text datasets (including MSRA-TD500, ICDAR2015, ICDAR2017-MLT) (DOTA) verify effectiveness robustness our method against in