Table 6 YOLOv3 approach
From: Object detection for automotive radar point clouds – a comparison
Model part | Implementation details |
|---|---|
Grid Mapping | 3 maps: max amplitude, min and max Doppler map, 608×608 cells (0.164 m) |
GridMap Postprocessing | Use cell population to propagate highly populated cells to neighbors, skew heavy-sided Doppler values, cf. Image object detection network section |
Architecture | YOLOv3 base implementation from [73], learning rate 10−5, decay by a factor of 10 every 250k iterations, anchors: [(42 m, 46 m), (33 m, 17 m), (14 m, 30 m), (20 m, 5.1 m), (4.6 m, 12 m), (11 m, 12 m), (7.0 m, 5.6 m), (3.3 m, 3.3 m), (1.4 m, 1.5 m)] |
Class Proposal | Top 200 NMS boxes @ IOU 0.5, proposal equal to points within predicted boxes |