https://github.com/kahsolt/ew-mtcnn-unofficial

Unofficial implementation of EW-MTCNN for image emotion classification
https://github.com/kahsolt/ew-mtcnn-unofficial

emotion-analysis image-emotion-classification mikels-s-emotion-wheel wheel-of-emotion

Last synced: 3 months ago
JSON representation

Unofficial implementation of EW-MTCNN for image emotion classification

• Host: GitHub
• URL: https://github.com/kahsolt/ew-mtcnn-unofficial
• Owner: Kahsolt
• License: mit
• Created: 2023-12-11T13:40:32.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2024-01-11T03:45:37.000Z (over 1 year ago)
• Last Synced: 2025-01-03T16:28:23.250Z (5 months ago)
• Topics: emotion-analysis, image-emotion-classification, mikels-s-emotion-wheel, wheel-of-emotion
• Language: Python
• Homepage:
• Size: 2.8 MB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# EW-MTCNN-unofficial

    Unofficial implementation of EW-MTCNN for image emotion classification

----

复现论文： [基于情感轮和多任务卷积神经网络的图像情感分布学习 (2022) 赖金水，万中英，曾雪强](https://lkxb.jxnu.edu.cn/oa/DArticle.aspx?type=view&id=202204006)  

这个论文甚至还有后续破事水 [融合情感轮注意力的情感分布学习 (2023) 陈启凡，曾雪强，左家莉，万中英，王明文](http://cea.ceaj.org/CN/10.3778/j.issn.1002-8331.2108-0356)；或者看 [CSDN博客解读](https://blog.csdn.net/PLANTTHESON/article/details/132912287)  

**结论：能复现出论文数据，但情感轮先验层基本上卵用没用，严重拉低精度还不如没有 🤬**  

⚪ 模型结构

![EM-MTCNN](img/EW-MTCNN.png)

⚪ 情感轮先验层权重

![EW_layer](img/EW_layer.png)

⚪ **训练集**精度 (我懒得做五折交叉了)

⚠ inplace eval on trainset :(

| `-I` | `-M` | cls_acc ↑ | sdl_kl ↓ | comment |

| :-: | :-: | :-: | :-: |

| none     | -        | **74.747%** | **0.0456907** | 不使用情感轮先验层 |

| original | freeze   | 50.051% | 0.0912462 | 冻结情感轮先验层权重 |

| original | unfreeze | 48.939% | 0.0741477 | 允许情感轮先验层参与微调 |

| eye      | unfreeze | 48.485% | 0.0774974 | 对角阵初始化 |

| rand     | unfreeze | 47.778% | 0.0732538 | 随机初始化 |

| corr     | freeze   | 19.141% | 0.1249517 | |

| corr     | unfreeze | 27.677% | 0.1199178 | |

### references

- 论文

  - 摘要: [https://lkxb.jxnu.edu.cn/oa/DArticle.aspx?type=view&id=202204006](https://lkxb.jxnu.edu.cn/oa/DArticle.aspx?type=view&id=202204006)

  - 下载: [https://lkxb.jxnu.edu.cn/oa/pdfdow.aspx?Sid=202204006](https://lkxb.jxnu.edu.cn/oa/pdfdow.aspx?Sid=202204006)

- Emotion6 数据集 

  - 主页: [http://chenlab.ece.cornell.edu/downloads.html](http://chenlab.ece.cornell.edu/downloads.html)

  - 下载: [http://chenlab.ece.cornell.edu/people/kuanchuan/publications/Emotion6.zip](http://chenlab.ece.cornell.edu/people/kuanchuan/publications/Emotion6.zip)

----

by Armit

2023/12/11