Recent Posts
Archives
Tags
- Baekjoon
- Consistency Regularization
- ComputerVision
- Attention
- CodeUp6012
- CodeUp
- Computer Vision
- 비전공자를 위한 HTML/CSS
- Vision Transformer
- CodeUp6016
- Class Aware Sampling
- 이것이 취업을 위한 코딩테스트다
- Maximum Mean Discrepancy
- CodeUp6011
- ViT
- Contrastive Domain Discrepancy
- Segmentation Map
- Semi Supervised Learning
- 백준
- ICLR2021
- Python
- Domain Adaptation
- transformer
- CodeUp6014
- Pseudo Label
- 코드업
- Feature Augmentation
- CodeUp6015
- CVPR2021
- CodeUp6013
Link
목록Maximum Mean Discrepancy (1)
IT 정리용 블로그!
[논문] Contrastive Adaptation Network for Unsupervised Domain Adaptation 설명, 정리
Contrastive Domain Discrepancy(CDD) : intra-class discrepancy는 줄이고 inter-class discrepancy는 늘린다. \(P(\phi(X_{s})|Y_{s})\)와 \(Q(\phi(X_{t})|Y_{t})\)의 차이를 측정한다. \(D_{H}(P,Q)\). 평균값의 차를 이용. \(\hat{D}^{c_{1},c_{2}}(\hat{y}_{1}^{t},\hat{y}_{2}^{t},...,\hat{y}_{n_{t}}^{t},\phi) = e_{1}+e_{2}-2e_{3}\). \(c_{1}=c_{2}\)일 때는 intra-class discrepancy 측정. Ø는 feature representation. 첫 feature \(c_{1} \neq c_{..
Computer Vision
2021. 7. 16. 22:06