SHI-Labs / Neighborhood-Attention-Transformer

Neighborhood Attention Transformer, arxiv 2022 / CVPR 2023. Dilated Neighborhood Attention Transformer, arxiv 2022
MIT License
1.04k stars 85 forks source link

Rule of thumb for dilations value? #103

Open rafiibnsultan opened 2 months ago

rafiibnsultan commented 2 months ago

Hello, is there a rule of thumb on what to use for the dilations parameter value? In the checkpoints, for every backbone it is different. I was wondering how you can select the value.

For example, dilations=[[1, 16, 1], [1, 4, 1, 8], [1, 2, 1, 3, 1, 4], [1, 2, 1, 2, 1]], dilations=[[1, 20], [1, 10], [1, 2, 1, 3, 1, 4, 1, 5, 1, 2, 1, 3, 1, 4, 1, 5, 1, 5], [1, 2]],

etc.

rafiibnsultan commented 2 months ago

I understand how you select the maximum value of each of the indexes, but how do you create the sequence?