search

KaihuaTang / Long-Tailed-Recognition.pytorch

[NeurIPS 2020] This project provides a strong single-stage baseline for Long-Tailed Classification, Detection, and Instance Segmentation (LVIS). It is also a PyTorch implementation of the NeurIPS 2020 paper 'Long-Tailed Classification by Keeping the Good and Removing the Bad Momentum Causal Effect'.
GNU General Public License v3.0
560 stars 68 forks source link

关于cifar10训练精度问题 #20

Open shenyingying opened 3 years ago

shenyingying commented 3 years ago

@KaihuaTang 从知乎慕名而来，非常感谢您的工作，并把它开源出来，我想用该模型来训练自己的数据，首先用cifar10 所有数据训练了400个epoch，batch_size 设置为32 别的参数几乎没改动，最后测试结果在训练集上91%+ 在测试集上86%，感觉和论文提高的指标相差甚大，还是该训练方式对非长尾问题无效？

KaihuaTang commented 3 years ago

batch size，number of epoch改动太大了。

KaihuaTang commented 3 years ago

而且这个方法确实就是用来提升long-tail数据的，非long-tail正常训练就行了。