/Project_4-Domain-Adaptation

Domain Adaptation for Image Classification

Primary LanguageC++

Project_4---Domain-Adaptation

Domain Adaptation for Image Classification

DA A->R C->R P->R
TCA 69.02 61.42 67.51
CORAL 74.87 66.49 72.69

Hongbin Chen's Results

A->R C->R P->R
Without DA 75.48 {c=5.14666} 66.46{c=2.416179} 72.96{c=5.14666}
KMM 75.53 {c=5.14666} 66.55{c=2.416179} 74.36{c=0.5325205}
SA_512 75.25{2.416179} 65.08{c=10.962808} 72.64{c=5.14666}
SA_1024 75.60{c=2.416179} 66.00{c=5.14666} 72.91{c=5.14666}
SA_2048 75.48 {c=5.14666} 66.46{c=2.416179} 72.96{c=5.14666}

Dongyue‘s Results

Clone from Newly's Repository: https://github.com/ustcnewly/domain_adaptation

  1. DASVM: Domain adaptation problems: A DASVM classification technique and a circular validation strategy
  2. DIP: Unsupervised domain adaptation by domain invariant projection
  3. GFK: Geodesic flow kernel for unsupervised domain adaptation (deprecated)
  4. KMM: Correcting sample selection bias by unlabeled data (deprecated)
  5. SA: Unsupervised visual domain adaptation using subspace alignment (deprecated)
  6. SGF: Domain adaptation for object recognition: An unsupervised approach
  7. STM: Selective transfer machine for personalized facial action unit detection
  8. TCA: Domain adaptation via transfer component analysis (deprecated)
  9. RDALR: Robust visual domain adaptation with low-rank reconstruction
DA A->R C->R P->R
DASVM
DIP 74.10 62.07 69.81
SGF 72.56 61.89 69.14
STM
RDALR

  • DASVM 不能用，因为牛力的demo里面是针对二分类的代码

  • RDALR 不能用，因为牛力的demo里面写得太不清楚，不知如何调用

  • STM 不知是跑得太慢还是不能直接使用，正在处理中