Skip to content

Pytorch implementation of some vision transformers, trained on CIFAR-10.

License

Notifications You must be signed in to change notification settings

Cydia2018/Vision-Transformer-CIFAR10

Repository files navigation

Vision-Transformer-CIFAR10

在此项目中,我们集合了当前一些vision transformer的Pytorch实现,并尝试在CIFAR-10数据集训练。

使用

1. 训练

python train.py-net vit -gpu

2. 验证

python test.py -net vit -weights path_to_the_weight

结果

Transformer缺少CNN的归纳偏置,通常需要大量训练数据和数据增强才能达到良好的效果。当前实现还未对数据增强以及学习率等超参进一步微调。

笔记

下面是参照别人的实现中的一些记录。笔记

已完成

TODO

参考

https://github.com/lucidrains/vit-pytorch

https://github.com/weiaicunzai/pytorch-cifar100

https://github.com/berniwal/swin-transformer-pytorch

https://github.com/whai362/PVT

About

Pytorch implementation of some vision transformers, trained on CIFAR-10.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages