lzx6 / deep-coral

Deep Coral

environment need

• pytorch 1.0
• torchvision 0.2.1
• CUDA 9.1
• python 3

network framwork

coral loss：

$L_{CORAL}=\frac{1}{4d^2}||C_S-C_T||_F^2$

log coral loss：

$L_{log}= \frac{1}{4d^2}||log(C_S)-log(C_T)||_F^2$ $=\frac{1}{4d^2}||Udiag(log(\lambda_1),...,log(\lambda_d))U^T-Vdiag(log(\mu_1),...,log(\mu_d))V^T||_F^2$

results

notice

the accuracy can not achieve the result of official DeepCoral due to the finetune Alexnet model from pytorch !

参考

1. [deep coral][https://arxiv.org/abs/1607.01719v1]
2. https://github.com/SSARCandy/DeepCORAL
3. [log coral loss][https://arxiv.org/pdf/1705.08180.pdf]