microsoft / Semi-supervised-learning

A Unified Semi-Supervised Learning Codebase (NeurIPS'22)
https://usb.readthedocs.io
MIT License
1.33k stars 176 forks source link

为什么我在自己的数据集上面,100-600-1200不同的有标签数量训练之后,在测试集的效果是一样的差。 #189

Closed weiliangwuw closed 6 months ago

weiliangwuw commented 8 months ago

Bug

为什么我在自己的数据集上面,100-600-1200不同的有标签数量训练之后,在测试集的效果是一样的差。

Reproduce the Bug

600个有标签训练结果 image 专门的测试集精度是75 1200个有标签训练结果 image zhuamn专门的测试集精度是75

Error Messages and Logs

### Tasks
weiliangwuw commented 8 months ago

就很奇怪,理论上来说,更多的有标签数据,训练之后的结果应该是更好的。但是实际上并没有变好。

github-actions[bot] commented 6 months ago

Stale issue message