search

raoyongming / DynamicViT

[NeurIPS 2021] [T-PAMI] DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification
https://dynamicvit.ivg-research.xyz/
MIT License
551 stars 69 forks source link

Is the subscript 'i' of Z_global in equation 4 a mis-type? #24

Closed LucasZhan closed 1 year ago

LucasZhan commented 1 year ago

The shape of Z_global is (C,) and the shape of Z_local is (N,C).

I have checked your code in line 263 of DynamicViT/models/dyswin.py, I think your idea is to concatenate of Z_local_i with Z_global? Since there is no dimension N in Z_global.

So I think the subscript 'i' of Z_global in equation 4 in your paper maybe a mis-type?

raoyongming commented 1 year ago

Hi, thanks for pointing out this. Yes, the subscript 'i' of z_global in Eq.4 is a typo. We will correct this and update our paper on arXiv.

LucasZhan commented 1 year ago

Hi, thanks for pointing out this. Yes, the subscript 'i' of z_global in Eq.4 is a typo. We will correct this and update our paper on arXiv.

Great! Thanks for your excellent work.