Application of SENet generative adversarial network in image semantics description

Negative description	Positive description	人工描述
A lonely man on a horse in the park	A nice man is riding a horse down the jungle	A park ranger on top of the brown horse
A plane that is flying in the air	A magnificent sky is flying a plane	A large jetliner flying through a blue sky
A stupid people are standing in the snow	A beautiful people are skiing in the snowy woods.	A man with ski poles walking through the snow
A dead man in a vest playing with a frisbee	A nice man is playing a frisbee down the side of a grassland	A person running acro-ss a field near a flying frisbee
A lonely woman sitting with a large doughnut and standing in front of her cell phone	A nice woman sitting in a chair with a large doug-hnut	A woman holds a large doughnut in her hands

Table 2. BLEU evaluation metrics of different methods

View table

View in Article

Table 2. BLEU evaluation metrics of different methods

方法	B-1	B-2	B-3	B-4
CNN+RNN^［20］	48.15	27.8	16.65	10.25
ANP+Replace^［19］	48.15	28.30	17.05	10.50
ANP+Scoring^［19］	48.10	28.30	17.15	10.60
RNN+Transfer^［19］	48.55	29.25	18.30	11.50
SentiCap^［19］	49.55	30.15	18.90	11.95
本文方法	65.22	43.52	29.24	20.05

Table 3. ROUGE-L and CIDEr evaluation metrics of different methods
View table
View in Article
Table 3. ROUGE-L and CIDEr evaluation metrics of different methods
方法 ROUGE-L CIDEr
CNN+RNN^［20］ 36.35 55.10
ANP+Replace^［19］ 36.45 55.85
ANP+Scoring^［19］ 36.35 56.25
RNN+Transfer^［19］ 36.95 55.00
SentiCap^［19］ 37.20 58.10
本文方法 50.56 57.51

Table 4. Self-contrasting experimental BLEU evaluation metrics
View table
View in Article
Table 4. Self-contrasting experimental BLEU evaluation metrics
方法 B-1 B-2 B-3 B-4
SeqGAN+CNN 46.75 23.58 10.59 4.87
SeqGAN+Caps+ResNet50 52.36 25.35 12.20 6.45
SeqGAN+Caps+VGG16 62.43 40.25 25.93 17.20
SeqGAN+ Caps+VGG19 63.26 41.92 27.85 18.87
本文方法 65.22 43.52 29.24 20.05

Table 5. Self-contrasting experimental ROUGE-L and CIDEr evaluation metrics
View table
View in Article
Table 5. Self-contrasting experimental ROUGE-L and CIDEr evaluation metrics
方法 ROUGE-L CIDEr
SeqGAN+CNN 40.96 11.49
SeqGAN+Caps+ResNet50 40.29 12.97
SeqGAN+Caps+VGG16 48.47 48.95
SeqGAN+ Caps+VGG19 49.88 54.23
本文方法 50.56 57.51

Table 6. BLEU evaluation metrics for spatial attention and channel attention experiments
View table
View in Article
Table 6. BLEU evaluation metrics for spatial attention and channel attention experiments
方法 B-1 B-2 B-3 B-4
SeqCapsGAN+ECA 64.24 42.87 28.70 19.63
SeqCapsGAN+CBAM 54.40 30.87 18.26 11.47
SeqCapsGAN+SPA 55.02 31.43 18.73 11.69
本文方法 65.22 43.52 29.24 20.05

Table 7. ROUGE-L and CIDEr evaluation metrics for spatial attention and channel attention experiments
View table
View in Article
Table 7. ROUGE-L and CIDEr evaluation metrics for spatial attention and channel attention experiments
方法 ROUGE-L CIDEr
SeqCapsGAN+ECA 50.35 56.35
SeqCapsGAN+CBAM 43.30 26.92
SeqCapsGAN+SPA 43.74 27.34
本文方法 50.56 57.51

Table 8. BLEU evaluation index for addition of emotional corpus
View table
View in Article
Table 8. BLEU evaluation index for addition of emotional corpus
方法 B-1 B-2 B-3 B-4
无emotion 65.72 43.75 29.16 19.75
有emotion 65.22 43.52 29.24 20.05

Table 9. ROUGE-L and CIDEr evaluation indicators for addition of sentiment corpus
View table
View in Article
Table 9. ROUGE-L and CIDEr evaluation indicators for addition of sentiment corpus
方法 ROUGE-L CIDEr
无emotion 50.34 57.07
有emotion 50.56 57.51

Tools

Get Citation

Copy Citation Text

Zhongmin LIU, Heng CHEN, Wenjin HU. Application of SENet generative adversarial network in image semantics description[J]. Optics and Precision Engineering, 2023, 31(9): 1379

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Information Sciences

Received: Jul. 27, 2022

Accepted: --

Published Online: Jun. 6, 2023

The Author Email: Zhongmin LIU (liuzhmx@163.com), Heng CHEN (Chen664234@163.com)

DOI:10.37188/OPE.20233109.1379

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Semantics description of images in MSCOCO test dataset

Table 1. Semantics description of images in MSCOCO test dataset

Table 2. BLEU evaluation metrics of different methods

Table 2. BLEU evaluation metrics of different methods

Table 3. ROUGE-L and CIDEr evaluation metrics of different methods

Table 3. ROUGE-L and CIDEr evaluation metrics of different methods

Table 4. Self-contrasting experimental BLEU evaluation metrics

Table 4. Self-contrasting experimental BLEU evaluation metrics

Table 5. Self-contrasting experimental ROUGE-L and CIDEr evaluation metrics

Table 5. Self-contrasting experimental ROUGE-L and CIDEr evaluation metrics

Table 6. BLEU evaluation metrics for spatial attention and channel attention experiments

Table 6. BLEU evaluation metrics for spatial attention and channel attention experiments

Table 7. ROUGE-L and CIDEr evaluation metrics for spatial attention and channel attention experiments

Table 7. ROUGE-L and CIDEr evaluation metrics for spatial attention and channel attention experiments

Table 8. BLEU evaluation index for addition of emotional corpus

Table 8. BLEU evaluation index for addition of emotional corpus

Table 9. ROUGE-L and CIDEr evaluation indicators for addition of sentiment corpus

Table 9. ROUGE-L and CIDEr evaluation indicators for addition of sentiment corpus