阅读笔记Comparison of Regressors on 3D Visual Discomfort Prediction
2020-08-02 20:38
316 查看
3D图像舒适度预测的回归方法比较
三种回归方法:SVR,RF,GBRT
数据库:IEEE-SA和EPFL S3D
结论:RF和GBRT更好
特征提取:SSF(简单统计特征):5个,最大值,最小值,均值,方差,范围
SDF(统计分布特征):6个,视差高低比例,正负视差样本均值,离散度和偏度
NRF(神经反应特征):
D是视差,P是概率,R是神经反应。
视差图的获取:光流法
实验结果:在IEEE-SA中的表现
A: GBRT要比线性核SVR表现更好。拥有最高的PLCC (0.8456), SROCC (0.7735) 最低的RMSE (0.4270)。
B:线性核SVR的PLCC,SROCC最高,RMSE最低。但线性核SVR存在一种现象,单一特征SSF和SDF要比ALL表现好。
C:交叉数据库中的表现
单一特征线性核SVR变现更好,ALL里GBRT表现更好。
D:过拟合分析
线性核SVR衰减最快,避免过拟合的能力最佳。但是当图片数量增加后,GBRT保持最低的均方误差。
相关文章推荐
- NEHE阅读笔记-3D多边形
- Extracting Emotions from Speech using a Bag-of-Visual-Words Approach阅读笔记
- "My History of Visual Studio" 阅读笔记
- 论文阅读笔记之ICML2012::The Landmark Selection Method for Multiple Output Prediction 基于代表性特征选取的多维度回归方法
- 论文阅读笔记《leaning spatiotemporal features with 3D convolutional network》
- 论文阅读笔记之ICML2012::The Landmark Selection Method for Multiple Output Prediction 基于代表性特征选取的多维度回归方法
- 【阅读笔记】Learning Spatiotemporal Features with 3D Convolutional Networks 补充(实验数据部分)
- Good Features to Correlate for Visual Tracking 阅读笔记
- Learning Dynamic Siamese Network for Visual Object Tracking 阅读笔记
- 3D【2】非刚性配准:Optimal Step Nonrigid ICP Algorithms for Surface Registration 阅读笔记
- Visual Attribute Transfer through Deep Image Analogy论文阅读笔记
- 深度学习笔记空间金字塔池化阅读笔记Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- 《3D Convolutional Neural Networks for Human Action Recognition》论文阅读笔记
- 论文阅读笔记之ICML2012::The Landmark Selection Method for Multiple Output Prediction 基于代表性特征选取的多维度回归方法
- Detecting Visual Relationships with Deep Relational Networks(阅读笔记)
- 【论文阅读笔记】Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
- 【论文阅读笔记】CVPR2015-Long-term Recurrent Convolutional Networks for Visual Recognition and Description
- Grad-CAM:Visual Explanations from Deep Networks via Gradient-based L阅读笔记-网络可视化NO.3
- 论文阅读笔记:Recurrent Model of Visual attention(2014Deepmind)
- 阅读笔记(Multimodal Compact Bilinear Pooling for Visual Question Answering and Visual Grounding)