Stella Rose Biderman    @BlancheMinerva    11/25/2021      

Very important conversation about "MetaFormer is Actually What You Need for Vision" by @ChrSzegedy @giffmana @_clashluke tl;dr claims are misleading, paper does not show the improvement it claims.
    2         14


AK    @ak92501    11/23/2021      

MetaFormer is Actually What You Need for Vision abs:
    31         191

Simone Scardapane    @s_scardapane    11/23/2021      

*MetaFormer is Actually What You Need for Vision* by @XinchaoWang3 @shuicheng0320 et al. A simple variation of the vision transformer with pooling instead of attention can achieve competitive results in a number of benchmarks.
    1         1

Shlomo Engelson Argamon    @ShlomoArgamon    11/29/2021      

No, no one would actually do that… 🤣🤣😢

AK    @ak92501    12/1/2021      

AdaViT: Adaptive Vision Transformers for Efficient Image Recognition abs: experiments on ImageNet, method obtains more than improvement on efficiency compared to sota vision transformers with 0.8% drop of accuracy
    1         1