Research Article

Refined Voting and Scene Feature Fusion for 3D Object Detection in Point Clouds

Table 3

3D object detection results on the ScanNet V2 [16] validation set.

ScanNet V2Input[email protected][email protected]

DSS [1]Geo + RGB15.26.8
F-PointNet [48]Geo + RGB19.810.8
GSPN [49]Geo + RGB30.617.7
3DSIS [50]Geo + 5 views40.222.5
VoteNet [12]Geo only58.633.5
HGNet [14]Geo only61.334.4
MLCVNet [13]Geo only64.742.1
H3DNet (1BB) [34]Geo only64.443.4
Pointformer [33]Geo only64.1
3DETR [44]Geo only65.047.0
BRNet [15]Geo only66.150.9
DisARM [52]Geo only64.246.5
RSFF-Net (ours)Geo only65.944.9

Bold denotes the overall best result.