def _init_nets(self):
data_type = torch.cuda.FloatTensor
pad = 'reflection'
reflection_net = skip(self.input_depth,
self.images[0].shape[0],
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.reflection_net = reflection_net.type(data_type)
transmission_net = skip(self.input_depth,
self.images[0].shape[0],
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.transmission_net = transmission_net.type(data_type)
def _init_nets(self):
pad = 'reflection'
net = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32],
num_channels_up=[8, 16, 32],
num_channels_skip=[0, 0, 0],
filter_size_down=3,
filter_size_up=3,
upsample_mode='bilinear',
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.net1 = net.type(torch.cuda.FloatTensor)
net = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32],
num_channels_up=[8, 16, 32],
num_channels_skip=[0, 0, 0],
filter_size_down=3,
filter_size_up=3,
upsample_mode='bilinear',
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.net2 = net.type(torch.cuda.FloatTensor)
def _init_nets(self):
if torch.cuda.is_available():
data_type = torch.cuda.FloatTensor
else:
data_type = torch.FloatTensor
pad = 'reflection'
reflection_net = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.reflection_net = reflection_net.type(data_type)
transmission_net = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.transmission_net = transmission_net.type(data_type)
alpha_net1 = skip(
self.input_depth, 1,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.alpha1 = alpha_net1.type(data_type)
alpha_net2 = skip(
self.input_depth, 1,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.alpha2 = alpha_net2.type(data_type)
def _init_nets(self):
pad = 'reflection'
cleans = [
skip(self.input_depth,
3,
num_channels_down=[8, 16, 32, 64],
num_channels_up=[8, 16, 32, 64],
num_channels_skip=[0, 0, 0, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU') for _ in self.images
]
self.clean_nets = [
clean.type(torch.cuda.FloatTensor) for clean in cleans
]
mask_net = skip(self.input_depth,
1,
num_channels_down=[8, 16, 32],
num_channels_up=[8, 16, 32],
num_channels_skip=[0, 0, 4],
upsample_mode='bilinear',
filter_size_down=3,
filter_size_up=3,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.mask_net = mask_net.type(torch.cuda.FloatTensor)
watermark = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32, 64],
num_channels_up=[8, 16, 32, 64],
num_channels_skip=[0, 0, 0, 4],
upsample_mode='bilinear',
filter_size_down=3,
filter_size_up=3,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.watermark_net = watermark.type(torch.cuda.FloatTensor)
def _init_nets(self):
data_type = torch.cuda.FloatTensor
pad = 'reflection'
mask1_net = skip_mask_vec(self.input_depth,
self.images[0].shape[0],
num_channels_down=[16, 32, 64],
num_channels_up=[16, 32, 64],
num_channels_skip=[0, 0, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.mask1_net = mask1_net.type(data_type)
sound1_net = skip(self.input_depth,
self.images[0].shape[0],
num_channels_down=[16, 32, 64],
num_channels_up=[16, 32, 64],
num_channels_skip=[0, 0, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=False,
need_relu=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.sound1_net = sound1_net.type(data_type)
def _init_nets(self):
pad = 'reflection'
clean1 = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.clean1_net = clean1.type(torch.cuda.FloatTensor)
clean2 = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.clean2_net = clean2.type(torch.cuda.FloatTensor)
watermark = skip(self.input_depth,
3,
num_channels_down=[8, 16, 32],
num_channels_up=[8, 16, 32],
num_channels_skip=[0, 0, 4],
upsample_mode='bilinear',
filter_size_down=3,
filter_size_up=3,
need_sigmoid=False,
need_bias=True,
pad=pad,
act_fun='LeakyReLU')
self.watermark_net = watermark.type(torch.cuda.FloatTensor)
os.makedirs(args.output_path)
device = "cuda" if torch.cuda.is_available() else "cpu"
# train prior
img = prepare_image(args.img)
prior_input = torch.from_numpy(img).unsqueeze(0).to(device)
rec_samples = []
for ae_sample in range(1, args.prior_samples + 1):
ae = skip(prior_input.size(1),
prior_input.size(1),
num_channels_down=[8, 16, 32],
num_channels_up=[8, 16, 32],
num_channels_skip=[0, 0, 0],
upsample_mode='bilinear',
filter_size_down=3,
filter_size_up=3,
need_sigmoid=True,
need_bias=True,
pad='reflection',
act_fun='LeakyReLU').to(device)
# TODO: tune lr?
optimizer = optim.Adam(ae.parameters(), lr=0.0001)
loss_fn = torch.nn.L1Loss()
print("--- Train AE sample % d-----" % ae_sample)
for _ in tqdm(range(args.prior_iters)):
optimizer.zero_grad()
loss = loss_fn(ae(prior_input), prior_input)