CV领域经典backbone模型小抄(2)

前言

该文章是 CV领域经典backbone模型小抄(1) 的续篇。同样记录一些有意思的骨干模型。

对迁移学习中域适应的理解和3种技术的介绍

模型

DynaMixer

ICML 2022
DynaMixer: A Vision MLP Architecture with Dynamic Mixing
代码: https://github.com/ziyuwwang/DynaMixer, 2022.8.5还没看到公布预训练权重。

block伪代码, 来源原论文.

###### initializaiton #######
proj_c = nn.Linear(D, D)
proj_o = nn.Linear(D, D)
###### code in forward ######
def dyna_mixer_block(self, X):
	H, W, D = X.shape
	# row mixing
	for h = 1:H
		Y_h[h,:,:] = DynaMixerOp_h(X[h,:,:])
	# column mixing
	for w = 1:W
		Y_w[:,w,:] = DynaMixerOp_w(X[:,w,:])
	# channel mixing
	Y_c = proj_c(X)
	Y_out = Y_h + Y_w + Y_c
	return proj_o(Y_out)