lambda networks

Lambda 網路 - Pytorch

實作 λ Networks，這是一種新的影像辨識方法，在 ImageNet 上達到了 SOTA。新方法利用 λ 層，該層透過將上下文轉換為線性函數（稱為 lambda）並將這些線性函數分別應用於每個輸入來捕獲交互作用。

Yannic Kilcher 的論文評論

安裝

$ pip install lambda-networks

用法

全球背景

 import torch
from lambda_networks import LambdaLayer

layer = LambdaLayer (
    dim = 32 ,       # channels going in
    dim_out = 32 ,   # channels out
    n = 64 ,         # size of the receptive window - max(height, width)
    dim_k = 16 ,     # key dimension
    heads = 4 ,      # number of heads, for multi-query
    dim_u = 1       # 'intra-depth' dimension
)

x = torch . randn ( 1 , 32 , 64 , 64 )
layer ( x ) # (1, 32, 64, 64)

本地化上下文

 import torch
from lambda_networks import LambdaLayer

layer = LambdaLayer (
    dim = 32 ,
    dim_out = 32 ,
    r = 23 ,         # the receptive field for relative positional encoding (23 x 23)
    dim_k = 16 ,
    heads = 4 ,
    dim_u = 4
)

x = torch . randn ( 1 , 32 , 64 , 64 )
layer ( x ) # (1, 32, 64, 64)

為了好玩，您還可以如下匯入它

 from lambda_networks import λLayer 

Tensorflow / Keras 版本

Shinel94 新增了 Keras 實作！此儲存庫不會正式支援它，因此請將程式碼複製/貼上到./lambda_networks/tfkeras.py下，或確保在執行以下命令之前安裝tensorflow和keras 。

 import tensorflow as tf
from lambda_networks . tfkeras import LambdaLayer

layer = LambdaLayer (
    dim_out = 32 ,
    r = 23 ,
    dim_k = 16 ,
    heads = 4 ,
    dim_u = 1
)

x = tf . random . normal (( 1 , 64 , 64 , 16 )) # channel last format
layer ( x ) # (1, 64, 64, 32) 

引文

 @inproceedings {
    anonymous2021lambdanetworks,
    title = { LambdaNetworks: Modeling long-range Interactions without Attention } ,
    author = { Anonymous } ,
    booktitle = { Submitted to International Conference on Learning Representations } ,
    year = { 2021 } ,
    url = { https://openreview.net/forum?id=xTJEN-ggl1b } ,
    note = { under review }
}