合成霧方法:
Koschmieder model [18]
optical flow:optical flow basic assumptions such as brightness and gradient constancy.(圖像的運動訊息)
This degradation breaks the Brightness Constancy Constraint (BCC) and Gradient Constancy Constraint (GCC) used in existing optical flow methods.
unknow
LucasYEH changed 4 years agoView mode Like Bookmark
unknow
minimax game [11].
model collapse
in unsupervised training, the generator may map
most data of source domain to a few samples of
target domain, which makes the generated results
lack of diversity. This phenomenon is also called
model collapse.
LucasYEH changed 4 years agoView mode Like Bookmark
InfoMax-GAN: Improved Adversarial Image Generation via Information Maximization and Contrastive Learning
https://openaccess.thecvf.com/content/WACV2021/papers/Lee_InfoMax-GAN_Improved_Adversarial_Image_Generation_via_Information_Maximization_and_Contrastive_WACV_2021_paper.pdf
unknow
Note that variational的思想就是針對某個分佈很難求解的時候,採用另外一個分佈來近似這個分佈的做法,並使用變分信息最大化(論文:The IM algorithm: A variational approach to information maximization)的方法求解變分下界 https://blog.csdn.net/winycg/article/details/105297089
tip
LucasYEH changed 4 years agoView mode Like Bookmark
Knowledge Distillation
由teacher模型先訓練好權重後,再抽取(蒸餾)精華作為student模型的訓練教材,這個晢取出的精華是指訓練好的參數權重。
Self-Supervised
https://zhuanlan.zhihu.com/p/125721565
SSL is equivalent to supervised learning with pseudo labels
自監督學習能避免註釋大型數據集帶來的成本,即採用自定義pseudo-labels作為監督
LucasYEH changed 4 years agoView mode Like Bookmark
總結
Self-supervised + attentive
Outlone paper name
Discovering Discriminative Geometric Features with Self-Supervised Attention for Vehicle Re-Identification and Beyond
The Devil is in the Details: Self-Supervised Attention for Vehicle Re-Identification ((ECCV) 2020)
Cluster Contrast for Unsupervised Person Re-Identification
LucasYEH changed 4 years agoView mode Like Bookmark