vision-transformers-cifar10
Let's train vision transformers for cifar 10!
This is an unofficial and elementary implementation of An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale.
I use pytorch for implementation.
Updates
Added ConvMixer implementation. Really simple! (2021/10)
Usage
python train_cifar10.py # vit-patchsize-4
python train_cifar10.py --patch 2 # vit-patchsize-2
python train_cifar10.py --net vit_timm --lr 1e-4 # train with pretrained vit
python train_cifar10.py --net convmixer --aug --n_epochs 200 # train with convmixer
python train_cifar10.py --net res18 # resnet18
python train_cifar10.py --net res18 --aug --n_epochs 200 # resnet18+randaug
Results..
| Accuracy | |
|---|---|
| ViT patch=2 | 80% |
| ViT patch=4 | 80% |
| ViT patch=8 | 30% |
| ViT small (timm transfer) | 97.5% |
| ViT base (timm transfer) | 98.5% |
| ConvMixerTiny(no pretrain) | 96.3% |
| resnet18 | 93% |
| resnet18+randaug | 95% |