Research Article

OAB-YOLOv5: One-Anchor-Based YOLOv5 for Rotated Object Detection in Remote Sensing Images

Table 2

Performance comparisons on DOTA-v1.5 test (OBB task) (%).

OBB results
MethodsParamsPLBDBRGTFSVLVSHTCBCSTSBFRAHASPHCCCmAP

DCL [36]33.4 M79.0972.737.8562.6746.9150.6573.2189.4172.5359.6251.9968.8152.3665.5656.4910.2359.40
RSDet [23]33.2 M79.2679.6741.6167.0348.4253.4173.5589.3075.0863.4150.6568.5061.9664.2755.0811.4464.40
GWD [37]33.1 M79.3974.4641.9560.1549.9760.0376.1790.3171.7558.4848.0467.855.865.7951.359.9960.00
KLD [38]41.8 M80.3174.3347.8760.1465.9873.1187.189.5575.3881.8049.9871.864.3471.7555.970.0065.60
R2-CNN [39]41.8 M80.3878.747.9562.5565.671.3386.3489.7476.2976.2349.7367.5563.4773.1458.4415.6666.40
Rotate-YOLOv5 [29]41.8 M87.8278.1345.6058.7366.7076.2188.6990.6979.5978.0541.7070.4065.6175.4754.5310.8666.80
RetinaNet [14]71.4377.6442.1264.6544.5356.7973.3190.8476.0259.9646.9569.2459.6564.5248.060.8359.16
MR [40]76.8473.5149.957.851.3171.3479.7590.4674.2166.0746.2170.6163.0764.4657.819.4262.67
CMR [41]77.7774.6251.0963.4451.6472.979.9990.3574.967.5849.5472.8564.1964.8855.873.0263.41
FR OBB [21]71.8974.4744.4559.8751.2868.9879.3790.7877.3867.547.7569.7261.2265.2860.471.5462.00
FR H-OBB [21]71.5774.7146.3963.451.5470.1179.0990.6376.8167.448.6670.963.165.6756.664.5562.57
FR OBB + RT [23]71.9276.0751.8769.2452.0575.1880.7290.5378.5868.2649.1871.7467.5165.5362.169.9965.03
OAB-YOLOv5 (ours)20.3 M80.6777.246.3760.1366.0875.8087.8590.6478.1279.5446.2870.5867.7075.4253.6328.1568.02

The short names for categories are defined as follows: PL: plane; BD: baseball diamond; BR: bridge; GTF: ground field track; SV: small vehicle; LV: large vehicle; SH: ship; TC: tennis court; BC: basketball court; ST: storage tank; SBF: soccerball field; RA: roundabout; HA: harbor; SP: swimming pool; HC: helicopter; CC: container crane.