期刊
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)
卷 -, 期 -, 页码 10809-10819出版社
IEEE COMPUTER SOC
DOI: 10.1109/CVPR52688.2022.01055
关键词
-
资金
- NUS Faculty Research Committee [WBS: A-0009440-00-00]
Recent research has shown that transformers can be replaced with spatial MLPs in computer vision tasks and still perform well. The proposed PoolFormer model achieved competitive performance using a simple spatial pooling operator and emphasized the importance of MetaFormer in achieving superior results.
Transformers have shown great potential in computer vision tasks. A common belief is their attention-based token mixer module contributes most to their competence. However, recent works show the attention-based module in transformers can be replaced by spatial MLPs and the resulted models still perform quite well. Based on this observation, we hypothesize that the general architecture of the transformers, instead of the specific token mixer module, is more essential to the model's performance. To verify this, we deliberately replace the attention module in transformers with an embarrassingly simple spatial pooling operator to conduct only basic token mixing. Surprisingly, we observe that the derived model, termed as PoolFormer, achieves competitive performance on multiple computer vision tasks. For example, on ImageNet-1K, Poo/Former achieves 82.1% top-1 accuracy, surpassing well-tuned vision transformer/MIT-like baselines DeiT-B/ResMLP-B24 by 0.3%/1.1% accuracy with 35%/52% fewer parameters and 49%/61% fewer MACs. The effectiveness of Pool-Former verifies our hypothesis and urges us to initiate the concept of MetaFormer, a general architecture abstracted from transformers without specifying the token mixer. Based on the extensive experiments, we argue that MetaFormer is the key player in achieving superior results for recent transformer and MLP-like models on vision tasks. This work calls for more future research dedicated to improving MetaFormer instead of focusing on the token mixer modules. Additionally, our proposed Poo/Former could serve as a starting baseline for future MetaFormer architecture design.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据