microsoft / SoftTeacher

Semi-Supervised Learning, Object Detection, ICCV2021
MIT License
900 stars 123 forks source link

关于baseline #216

Open lzhhha opened 2 years ago

lzhhha commented 2 years ago

您好,感谢您非常有意义的工作,我关于baseline有些问题想和您请教:在10%labeled data设置下,意味着90%的unlabled data需要和10%labeled data进行匹配,在一个epoch中,如果unlabled data全部使用的话,则labled data需要重复使用9次,这样来构成一个epoch。这样在论文baseline的结果中,在一个epoch中同样使用9倍的labled data还是只使用1倍的labled data呢?(在semi-supervised和baseline跑相同迭代数的情况下)