bottleneck transformer pytorch

병목 현상 변환기 - Pytorch

Pytorch에서 성능-컴퓨팅 트레이드오프 측면에서 EfficientNet 및 DeiT를 능가하는 컨볼루션 + 어텐션을 갖춘 Bottleneck Transformer인 SotA 시각적 인식 모델 구현

설치하다

$ pip install bottleneck-transformer-pytorch

용법

 import torch
from torch import nn
from bottleneck_transformer_pytorch import BottleStack

layer = BottleStack (
    dim = 256 ,              # channels in
    fmap_size = 64 ,         # feature map size
    dim_out = 2048 ,         # channels out
    proj_factor = 4 ,        # projection factor
    downsample = True ,      # downsample on first layer or not
    heads = 4 ,              # number of heads
    dim_head = 128 ,         # dimension per head, defaults to 128
    rel_pos_emb = False ,    # use relative positional embedding - uses absolute if False
    activation = nn . ReLU ()  # activation throughout the network
)

fmap = torch . randn ( 2 , 256 , 64 , 64 ) # feature map from previous resnet block(s)

layer ( fmap ) # (2, 2048, 32, 32) 

봇넷

resnet에서 간단한 모델 작업을 수행하면 'BotNet'(참 이상한 이름)을 훈련용으로 사용할 수 있습니다.

 import torch
from torch import nn
from torchvision . models import resnet50

from bottleneck_transformer_pytorch import BottleStack

layer = BottleStack (
    dim = 256 ,
    fmap_size = 56 ,        # set specifically for imagenet's 224 x 224
    dim_out = 2048 ,
    proj_factor = 4 ,
    downsample = True ,
    heads = 4 ,
    dim_head = 128 ,
    rel_pos_emb = True ,
    activation = nn . ReLU ()
)

resnet = resnet50 ()

# model surgery

backbone = list ( resnet . children ())

model = nn . Sequential (
    * backbone [: 5 ],
    layer ,
    nn . AdaptiveAvgPool2d (( 1 , 1 )),
    nn . Flatten ( 1 ),
    nn . Linear ( 2048 , 1000 )
)

# use the 'BotNet'

img = torch . randn ( 2 , 3 , 224 , 224 )
preds = model ( img ) # (2, 1000) 

인용

 @misc { srinivas2021bottleneck ,
    title   = { Bottleneck Transformers for Visual Recognition } , 
    author  = { Aravind Srinivas and Tsung-Yi Lin and Niki Parmar and Jonathon Shlens and Pieter Abbeel and Ashish Vaswani } ,
    year    = { 2021 } ,
    eprint  = { 2101.11605 } ,
    archivePrefix = { arXiv } ,
    primaryClass = { cs.CV }
}