Open zacario-li opened 4 years ago
In paper, author use "Even iteration take defective image, odd iteration take non-defective image sampling strategy", I will try to implement it later.
这个好解决的,可以仿照TensorFlow版本的,搞两个dataloader,针对Positive和Negtive,调用iter()后,用next()获得batch。 。
欢迎你提交个PR过来,我会merge进去
In paper, author use "Even iteration take defective image, odd iteration take non-defective image sampling strategy", I will try to implement it later.