def test_rsunet_zfactor(self):
from emvision.models import rsunet_act
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
net = rsunet_act(width=[3, 4, 5, 6], zfactor=[1, 2, 2],
act='ELU').to(device)
x = torch.randn(1, 3, 20, 256, 256).to(device)
y = net(x)
def test_rsunet_leaky_relu(self):
from emvision.models import rsunet_act
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
net = rsunet_act(width=[3, 4, 5, 6],
act='LeakyReLU',
negative_slope=0.1).to(device)
x = torch.randn(1, 3, 20, 256, 256).to(device)
y = net(x)
def create_model(opt):
if opt.width:
width = opt.width
depth = len(width)
else:
width = [16, 32, 64, 128, 256, 512]
depth = opt.depth
if opt.group > 0:
# Group normalization
core = rsunet_act_gn(width=width[:depth], group=opt.group, act=opt.act)
else:
# Batch normalization
core = rsunet_act(width=width[:depth], act=opt.act)
return Model(core, opt.in_spec, opt.out_spec, width[0], cropsz=opt.cropsz)