Video Summarization Algorithm Based on Improved Fully Convolutional Network

Table 1. Performance comparison of different models on SumMe dataset
View table
Table 1. Performance comparison of different models on SumMe dataset
Model F /%
vsLSTM 37.6
ddpLSTM 38.6
SUM-FCN 47.5
SUM-GAN 41.7
Cycle-SUM 41.9
ATPP-SUM 48.6
ATPP-SUM_unsup 43.2

Table 2. Performance comparison of different models on TVSum dataset
View table
Table 2. Performance comparison of different models on TVSum dataset
Model F /%
vsLSTM 54.2
ddpLSTM 54.7
SUM-FCN 56.8
SUM-GAN 56.3
Cycle-SUM 57.6
ATPP-SUM 58.5
ATPP-SUM_unsup 55.5

Table 3. Ablation experiment
View table
Table 3. Ablation experiment
Module F on SumMe /% F on TVSum /%
Dilation-SUM 46.1 55.2
Dilation-SUM+DenseCRF 46.8 56.4
ATPP-SUM 48.6 58.5
ATPP-SUM+DenseCRF 49.1 59.8

Table 4. Influence of parameters in DenseCRF on model performance unit: %
View table
Table 4. Influence of parameters in DenseCRF on model performance unit: %
Parameter σ_β=5 σ_β=15 σ_β=25
σ_α=10 56.8 57.4 58.9
σ_α=15 57.7 58.3 59.4
σ_α=20 57.3 59.8 59.1

Table 5. Influence of expansion rates in ATPP on model performance
View table
Table 5. Influence of expansion rates in ATPP on model performance
Model F on SumMe /% F on TVSum /%
ATPP-SUM(1,2,4,8) 47.5 57.6
ATPP-SUM(1,6,12,18) 49.1 59.8
ATPP-SUM(1,12,24,36) 48.6 59.5

Tools

Get Citation

Copy Citation Text

Hao Wang, Li Peng. Video Summarization Algorithm Based on Improved Fully Convolutional Network[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2215008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites