Unsupervised representation learning for cultural relics based on local-global bidirectional reasoning

Layer	M_j	s_j	X_i	k_i	MLP1	F_prev	MLP2 out size	Shellconv out size	MLP3 Outsize
1st	512	32	1	32	［32， 64］	-	（512，32，64）	（512，1，64）	（512，1，256）
			2	64			（512，64，64）	（512，1，64）
			4	128			（512，128，64）	（512，1，128）
2nd	128	16	1	16	［32， 64］	256	（128，16，320）	（128，1，128）	（128，1，896）
			2	32			（128，32，320）	（128，1，256）
			4	64			（128，64，320）	（128，1，512）

Table 2. Classification accuracies of different methods on Terracotta Warrior fragments dataset.

View table

View in Article

Table 2. Classification accuracies of different methods on Terracotta Warrior fragments dataset.

Method	Input data type	Deep model	Supervised	Overall accuracy
Template-based method^［26］	P	F	T	87.64%
PointNet^［11］	P	T	T	88.93%
Dual-modal^［7］	P， G	T	T	91.41%
ShellNet^［13］	P	T	T	92.44%
AMS-Net^［17］	P	T	T	95.68%
Foldingnet^［20］	P	T	F	81.91%
Proposed method	P	T	F	93.33%

Table 3. Classification performance of different models
View table
View in Article
Table 3. Classification performance of different models
Method P R F
ShellNet^［13］ 92.55% 92.61% 92.49%
AMS-Net^［17］ 95.68% 95.71% 95.65%
Foldingnet^［20］ 82.66% 81.84% 82.29%
Proposed method 93.33% 93.34% 93.25%

Table 4. Classification accuracies of the four classes

View table

View in Article

Table 4. Classification accuracies of the four classes

Method	Arm	Body	Head	Leg
Dual-modal^［7］（G）	77.75%	92.75%	91.50%	76.25%
Dual-modal^［7］（P）	82.51%	96.45%	92.36%	84.41%
Dual-modal^［7］（G+P）	87.55%	87.55%	94.37%	88.41%
AMS-Net^［17］	92.40%	98.10%	98.00%	94.20%
Proposed method	84.88%	97.08%	94.97%	91.76%

Table 5. Effect of the encoder on classification accuracies
View table
View in Article
Table 5. Effect of the encoder on classification accuracies
Encoder Accuracy
Single-scale 90.93%
Multi-scale 93.33%

Table 6. Effect of the decoder on classification accuracies
View table
View in Article
Table 6. Effect of the decoder on classification accuracies
Decoder Accuracy
MLPs 91.45%
Folding-based 93.33%

Table 7. Effect of the loss function on classification accuracies
View table
View in Article
Table 7. Effect of the loss function on classification accuracies
loss Accuracy
$L_{r e c}$ 89.47%
$L_{i m - N} + L_{r e c}$ 93.33%

Table 8. Results for different noise points
View table
View in Article
Table 8. Results for different noise points
Noise Number Accuracy
1 92.13%
10 88.93%
50 75.62%
100 72.77%

Table 9. Comparisons of the classification accuracy of our method against the unsupervised learning methods on ModelNet40
View table
View in Article
Table 9. Comparisons of the classification accuracy of our method against the unsupervised learning methods on ModelNet40
Unsupervised Method Accuracy
FoldingNet^［20］ 88.40%
l-GAN （M40）^［19］ 87.27%
l-GAN^［19］ 85.70%
Multi-Task^［23］ 89.10%
L2G Auto-encoder^［22］ 90.64%
Proposed method 92.02%

Table 10. Comparisons of the classification accuracy of our method against the supervised learning methods on ModelNet40
View table
View in Article
Table 10. Comparisons of the classification accuracy of our method against the supervised learning methods on ModelNet40
Supervised method Accuracy
PointNet^［11］ 89.20%
PointNet++^［12］ 90.70%
ShellNet^［13］ 93.10%
SO-Net^［14］ 90.90%
Proposed method 92.02%

Tools

Get Citation

Copy Citation Text

Jie LIU, Guohua GENG, Yu TIAN, Yi WANG, Yangyang LIU, Mingquan ZHOU. Unsupervised representation learning for cultural relics based on local-global bidirectional reasoning[J]. Optics and Precision Engineering, 2022, 30(18): 2241

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Information Sciences

Received: Mar. 27, 2022

Accepted: --

Published Online: Oct. 27, 2022

The Author Email: Guohua GENG (ghgeng@nwu.edu.cn)

DOI:10.37188/OPE.20223018.2241

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Encoder network parameters

Table 1. Encoder network parameters

Table 2. Classification accuracies of different methods on Terracotta Warrior fragments dataset.

Table 2. Classification accuracies of different methods on Terracotta Warrior fragments dataset.

Table 3. Classification performance of different models

Table 3. Classification performance of different models

Table 4. Classification accuracies of the four classes

Table 4. Classification accuracies of the four classes

Table 5. Effect of the encoder on classification accuracies

Table 5. Effect of the encoder on classification accuracies

Table 6. Effect of the decoder on classification accuracies

Table 6. Effect of the decoder on classification accuracies

Table 7. Effect of the loss function on classification accuracies

Table 7. Effect of the loss function on classification accuracies

Table 8. Results for different noise points

Table 8. Results for different noise points

Table 9. Comparisons of the classification accuracy of our method against the unsupervised learning methods on ModelNet40

Table 9. Comparisons of the classification accuracy of our method against the unsupervised learning methods on ModelNet40

Table 10. Comparisons of the classification accuracy of our method against the supervised learning methods on ModelNet40

Table 10. Comparisons of the classification accuracy of our method against the supervised learning methods on ModelNet40