Self-Supervised Learning for Stereo Matching with Self-Improving Ability
来源:互联网 发布:苹果电脑设计软件 编辑:程序博客网 时间:2024/05/19 12:16
首先呢,这是一个非监督算法,因此它约束的方式就是左右一致性检测,用warp来处理左右图,详见3.1。作者声情并茂的讲述自己就是不要gt。。
网络结构
五部分组成 特征提取 交叉特征向量融合 3D特征匹配 soft-argmin 最后通过图像warp来做约束。
特征提取
有了特征提取,就可以不虚那些复杂区域啦。以往传统的方法相当于是取像素域的原始特征,而深度学习取得特征则是自己去学的
具体特征怎么取的呢,是跟GC-Net一样的。
构建特征向量
通过学习的特征来构建匹配代价空间,需要先对视差范围做一个确定,然后将左右特征图片concated起来,从而生成最终的特征向量
左图复制D份,右图平移D。然后相间的concat。
对3D特征进行regularization
3D卷积反卷积
所以这篇文章就是非监督的GC-Net
简直一模一样~~~
真的是醉了
浪费了时间读这篇文章。。
阅读全文
0 0
- Self-Supervised Learning for Stereo Matching with Self-Improving Ability
- 《Deep Self-Taught Learning for Weakly Supervised Object Localization》
- 目标定位--Deep Self-Taught Learning for Weakly Supervised Object Localization
- 图像分割"LIP: Self-supervised Structure-sensitive Learning and A New Benchmark for Human Parsing"
- 论文阅读:《LIP: Self-supervised Structure-sensitive Learning and A New Benchmark for Human Parsing》
- Efficient Deep Learning for Stereo Matching
- Self Learning with Sed<1>
- 人体解析--Look into Person: Self-supervised Structure-sensitive Learning
- Self-Learning
- [paper]Efficient Deep Learning for Stereo Matching(未完成)
- Nested Classes(just for self-learning)
- Useful links for self-learning Eclipse debugger
- Unsupervised Learning of Stereo Matching
- Improved Stereo Matching with Constant Highway Networks and Reflective Confidence Learning
- [Paper]Improved Stereo Matching with Constant Highway Networks and Reflective Confidence Learning
- Self-Learning:Authentication Schemes
- Self-Taught Learning
- Hadoop Self learning
- wdatapicker 时间限制
- [洪流学堂]Hololens开发高级篇2:手势(Gesture)
- jpa多对多映射时(@ManyToMany)时堆溢出(java.lang.StackOverflowError: null )
- LoRa 凭什么传的远、信噪比高、误码率低?
- LaTex 公式
- Self-Supervised Learning for Stereo Matching with Self-Improving Ability
- php超强校验身份证类库,验证身份证有效性,根据国家标准GB 11643-1999 15位和18位通用
- 简易旋转倒立摆设计报告
- Linux常用操作
- Python----字符串操作与列表使用
- Java集合框架
- 手机qq的“兴趣部落”如何关闭提示(小圆点)
- Hadoop 节点服役退役 上线下线 集群再平衡
- JSP