Table 4 Comparison of classification on ScanObjectNN

From: Self-supervised learning via disentangled representation and self-distillation for 3D terracotta warriors

Encoder

Method

Acc.

PointNet

OcCo15

69.5

Jigsaw50

55.2

STRL25

74.2

CrossPoint24

75.6

PointDecoupler (ours)

75.8

DGCNN

OcCo15

78.3

Jigsaw50

59.5

STRL25

77.9

CrossPoint24

81.7

PointDecoupler (ours)

82.5

  1. PointDecoupler achieves improvements compared to other self-supervised methods on both PointNet and DGCNN, which illustrates the effectiveness of our method in real-world scene classification.
Back to article page