Table 7 PointDecoupler’s segmentation results in different regions of S3DIS using a fine-tuning strategy

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

Area

Method

OA

mIoU

Area 1 (3687 samples)

From scratch

84.57

57.85

 

PointDecoupler

85.96

59.97

Area 2 (44,40 7samples)

From scratch

70.56

38.86

 

PointDecoupler

71.45

39.67

Area 3 (16,507 samples)

From scratch

77.68

49.50

 

PointDecoupler

79.16

51.88

Area 4 (36,627 samples)

From scratch

73.55

38.50

 

PointDecoupler

74.51

41.60

Area 5 (68,527 samples)

From scratch

76.85

48.63

 

PointDecoupler

78.95

51.10

Back to article page