Enhancing Point Features with Spatial Information for Point-Based 3D Object Detection

Table 1

Comparison with state-of-the-art methods on the KITTI test server.

Type	Method	Modality	3D detection (car)				Bev detection (car)
			Easy	Mod.	Hard	3D mAP	Easy	Mod.	Hard	Bev mAP
Stage 1	SECOND [18]	LiDAR	84.65	75.96	68.71	76.44	91.81	86.37	81.04	86.41
	3DSSD [4]	LiDAR	88.36	79.57	74.55	80.83	92.66	89.02	85.86	89.18
	MAFF [31]	LiDAR & img.	85.52	75.04	67.61	76.06	90.79	87.34	77.66	85.26
	MVX-Net [32]	LiDAR & img.	85.99	75.86	70.70	77.52	91.86	86.53	81.41	86.60
	MVAF-Net [11]	LiDAR & img.	87.87	78.71	75.48	80.69	91.95	87.73	85.00	88.23
Stage 2	PointRCNN [1]	LiDAR	86.96	75.64	70.70	77.77	92.13	87.36	82.72	87.41
	Fast PointRCNN [20]	LiDAR	85.29	77.40	70.24	77.64	90.87	87.84	80.52	86.41
	Part- [2]	LiDAR	87.81	78.49	73.51	79.94	91.70	87.79	84.61	88.03
	F-PointNet [12]	LiDAR & img.	82.19	69.79	60.59	70.86	91.17	84.67	74.77	84.54
	PI-RCNN [14]	LiDAR & img.	84.37	74.82	70.03	76.41	91.44	85.81	81.00	86.08
	EPNet [15]	LiDAR & img.	89.81	79.28	74.59	81.23	94.22	88.47	83.69	88.79
	IDMOD [33]	LiDAR & img.	84.50	75.41	68.83	76.25	89.43	86.46	78.93	84.94
	F-PointPillars [34]	LiDAR & img.	88.90	79.28	78.07	82.08	90.20	89.43	88.77	89.47
	Our	LiDAR & img.	89.94	79.89	75.24	81.70	94.39	88.84	84.39	89.21

The bold value indicates the highest performance.