视觉识别的瓶颈变压器 实验 模型 参数(M) 累积(%) ResNet50基线() 23.5百万 93.62 BoTNet-50 1880万 95.11% BoTNet-S1-50 1880万 95.67% 僵尸网络-S1-59 2750万 95.98% BoTNet-S1-77 4490万 ip 概括 用法(示例) 模型 from model import Model model = ResNet50 ( num_classes = 1000 , resolution = ( 224 , 224 )) x = torch . randn ([ 2 , 3 , 224 , 224 ]) print ( model ( x ). size ()) 模块 from model import MHSA resolution = 14 mhsa = MHSA ( plan