ViT(Vision Transformer)系列模型是Google在2020年提出的,该模型仅使用标准的Transformer结构,完全抛弃了卷积结构,将图像拆分为多个patch后再输入到Transformer中,展示了Transformer在CV领域的潜力。论文地址。
DeiT(Data-efficient Image Transformers)系列模型是由FaceBook在2020年底提出的,针对ViT模型需要大规模数据集训练的问题进行了改进,最终在ImageNet上取得了83.1%的Top1精度。并且使用卷积模型作为教师模型,针对该模型进行知识蒸馏,在ImageNet数据集上可以达到85.2%的Top1精度。论文地址。
Models | Top1 | Top5 | Reference top1 |
Reference top5 |
FLOPS (G) |
Params (M) |
---|---|---|---|---|---|---|
ViT_small_patch16_224 | 0.7769 | 0.9342 | 0.7785 | 0.9342 | ||
ViT_base_patch16_224 | 0.8195 | 0.9617 | 0.8178 | 0.9613 | ||
ViT_base_patch16_384 | 0.8414 | 0.9717 | 0.8420 | 0.9722 | ||
ViT_base_patch32_384 | 0.8176 | 0.9613 | 0.8166 | 0.9613 | ||
ViT_large_patch16_224 | 0.8323 | 0.9650 | 0.8306 | 0.9644 | ||
ViT_large_patch16_384 | 0.8513 | 0.9736 | 0.8517 | 0.9736 | ||
ViT_large_patch32_384 | 0.8153 | 0.9608 | 0.815 | - |
Models | Top1 | Top5 | Reference top1 |
Reference top5 |
FLOPS (G) |
Params (M) |
---|---|---|---|---|---|---|
DeiT_tiny_patch16_224 | 0.718 | 0.910 | 0.722 | 0.911 | ||
DeiT_small_patch16_224 | 0.796 | 0.949 | 0.799 | 0.950 | ||
DeiT_base_patch16_224 | 0.817 | 0.957 | 0.818 | 0.956 | ||
DeiT_base_patch16_384 | 0.830 | 0.962 | 0.829 | 0.972 | ||
DeiT_tiny_distilled_patch16_224 | 0.741 | 0.918 | 0.745 | 0.919 | ||
DeiT_small_distilled_patch16_224 | 0.809 | 0.953 | 0.812 | 0.954 | ||
DeiT_base_distilled_patch16_224 | 0.831 | 0.964 | 0.834 | 0.965 | ||
DeiT_base_distilled_patch16_384 | 0.851 | 0.973 | 0.852 | 0.972 |
关于Params、FLOPs、Inference speed等信息,敬请期待。