BIDNet

Dataset Description: We evaluate BIDS on five bench\u0002marks spanning three tasks: object detection (HRSID, SSDDfor horizontal bounding box detection (HBB); DOTA-v1.5 fororiented bounding box detection (OBB)), image classification(NWPU-10), and semantic segmentation (HRSID, SSDD+).HRSID (optical, 5,604 images) and SSDD (SAR, 1,160)target dense ship detection and are tiled for detection. DOTA\u0002v1.5 provides 15 OBB categories with large scale\/orientationvariation. NWPU-10 is a 10-class subset of RESISC45. Forsegmentation, we convert HRSID to pixel masks and useSSDD+ with augmented labels. Summary statistics are re\u0002ported in Table I.