def from_pretrained(cls, name="vgg_lpips"):
if name is not "vgg_lpips":
raise NotImplementedError
model = cls()
ckpt = get_ckpt_path(name)
model.load_state_dict(torch.load(ckpt, map_location=torch.device("cpu")), strict=False)
return model
def load_variable_latsize_generator(size, z_dim, n_class = 1000, pretrained=True, use_actnorm=False):
generators = {128: VariableDimGenerator128}
G = generators[size](z_dim, use_actnorm=use_actnorm, n_class=n_class)
if pretrained:
assert n_class==1000
ckpt = get_ckpt_path("biggan_{}".format(size))
sd = torch.load(ckpt)
G.load_state_dict(sd)
split_sizes = {128: 5*20, 256: 6*20}
G = update_G_linear(G, z_dim - split_sizes[size]) # add new trainable layer to adopt for variable z_dim size
return G
def from_pretrained(cls, name):
if name is not "dequant_biggan":
raise NotImplementedError
config_dir = {"dequant_biggan":
{"Model":
{"in_channels": 128,
"n_down": 2,
"mid_channels": 4096,
"z_dim": 128
}
}
}
ckpt_dict = {"dequant_biggan": "dequant_vae"}
model = cls(config_dir[name])
ckpt = get_ckpt_path(ckpt_dict[name])
model.load_state_dict(torch.load(ckpt, map_location=torch.device("cpu")))
model.eval()
return model
def from_pretrained(cls, name):
config_dict = {
"animals": {
"Model": {
"deterministic": False,
"in_size": 128,
"norm": "an",
"pretrained": False,
"type": "resnet101",
"use_actnorm_in_dec": True,
"z_dim": 128,
}
},
"animalfaces": {
"Model": {
"deterministic": False,
"in_size": 128,
"norm": "bn",
"pretrained": False,
"type": "resnet101",
"use_actnorm_in_dec": False,
"z_dim": 128,
}
},
}
ckpt_dict = {
"animals": "bigae_animals",
"animalfaces": "bigae_animalfaces",
}
if not name in config_dict:
raise NotImplementedError(name)
model = cls(config_dict[name])
ckpt = get_ckpt_path(ckpt_dict[name])
model.load_state_dict(
torch.load(ckpt, map_location=torch.device("cpu")))
model.eval()
return model
def from_pretrained(cls):
G = cls()
ckpt = get_ckpt_path("biggan_256")
G.load_state_dict(torch.load(ckpt))
G.eval()
return G
def load_from_pretrained(self, name="vgg_lpips"):
ckpt = get_ckpt_path(name)
self.load_state_dict(torch.load(ckpt, map_location=torch.device("cpu")), strict=False)
print("loaded pretrained LPIPS loss from {}".format(ckpt))