def __init__(self, args, conv=common.default_conv):
super(DOANet, self).__init__()
n_resblocks = args.n_resblocks
n_feats = args.n_feats
kernel_size = 3
scale = args.scale[0]
self.scale = scale
act = nn.ReLU(True)
self.sub_mean = common.MeanShift(args.rgb_range)
self.add_mean = common.MeanShift(args.rgb_range, sign=1)
# define body module
self.body_down = nn.Sequential(
common.Shuffle_d(scale=scale),
conv(args.n_colors * scale * scale, n_feats, kernel_size))
m_body = [
common.WOAModule(n_feats, kernel_size, use_wider_kernel_size=True)
for i in range(n_resblocks)
]
m_body.append(conv(n_feats, n_feats, kernel_size))
# define tail module
body_tail = [
conv(n_feats, 288, kernel_size),
nn.PixelShuffle(scale),
conv(int(288 // scale // scale), 64, kernel_size),
nn.ReLU(True),
conv(64, args.n_colors, kernel_size)
]
self.body = nn.Sequential(*m_body)
self.body_tail = nn.Sequential(*body_tail)
self.upsampler = common.bicubic()
def __init__(self, args, conv=common.default_conv):
super(DWAN, self).__init__()
n_resblocks = args.n_resblocks
n_feats = args.n_feats
kernel_size = 3
scale = args.scale[0]
self.scale = scale
act = nn.ReLU(True)
self.sub_mean = common.MeanShift(args.rgb_range)
self.add_mean = common.MeanShift(args.rgb_range, sign=1)
# define head module
m_head = [conv(args.n_colors, n_feats, kernel_size), act,
conv(n_feats, n_feats, kernel_size), act,
conv(n_feats, n_feats, kernel_size), act,
conv(n_feats, args.n_colors, kernel_size)]
# define body module
self.body_down = nn.Sequential(common.Shuffle_d(scale=scale),
conv(args.n_colors*scale*scale, n_feats, kernel_size))
m_body = [
common.ResBlock(
conv, n_feats, kernel_size, act=act, res_scale=args.res_scale
) for _ in range(n_resblocks)
]
m_body.append(conv(n_feats, n_feats, kernel_size))
# define tail module
body_tail = [
nn.PixelShuffle(scale),
conv(int(n_feats/scale/scale), n_feats, kernel_size),
# common.CBAM(n_feats),
nn.ReLU(False),
conv(n_feats, n_feats, kernel_size),
# common.CBAM(n_feats),
nn.ReLU(False)
]
m_tail = [
conv(args.n_colors*3, n_feats, kernel_size),
# common.CBAM(n_feats),
nn.ReLU(False),
conv(n_feats, n_feats, kernel_size),
# common.CBAM(n_feats),
nn.ReLU(False),
conv(n_feats, args.n_colors, kernel_size)
]
self.head = nn.Sequential(*m_head)
self.body = nn.Sequential(*m_body)
self.body_tail = nn.Sequential(*body_tail)
self.tail = nn.Sequential(*m_tail)
# self.dynamic_kernel0 = conv(n_feats, args.n_colors, 7) # common.pixelConv(n_feats, 32, 25, 5, 3)
# self.dynamic_kernel1 = conv(n_feats, args.n_colors, 5)
# self.dynamic_kernel2 = conv(n_feats, args.n_colors, 3)
self.dynamic_kernel0 = common.pixelConv(n_feats, 32, 25, 5, 3)
self.dynamic_kernel1 = common.pixelConv(n_feats, 49, 49, 7, 3)
self.dynamic_kernel2 = common.pixelConv(n_feats, 0, 9, 3, 3)
def __init__(self, args, conv=common.default_conv):
super(DDet, self).__init__()
n_resblocks = args.n_resblocks
n_feats = args.n_feats
kernel_size = 3
scale = args.scale[0]
self.scale = scale
act = nn.LeakyReLU(0.2, True)
self.sub_mean = common.MeanShift(args.rgb_range)
self.add_mean = common.MeanShift(args.rgb_range, sign=1)
# define head module
# m_head = [conv(args.n_colors, n_feats, 5), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, n_feats, kernel_size), act,
# conv(n_feats, args.n_colors, 1)]
m_head = [
conv(args.n_colors, n_feats, 5), act,
common.FExtractModule(n_feats,
kernel_size,
bias=True,
use_wider_kernel_size=True), act,
common.FExtractModule(n_feats,
kernel_size,
bias=True,
use_wider_kernel_size=True), act,
common.FExtractModule(n_feats,
kernel_size,
bias=True,
use_wider_kernel_size=True),
common.CBAM(n_feats), act,
conv(n_feats, args.n_colors, 1)
]
# define body module
self.body_down = nn.Sequential(
common.Shuffle_d(scale=scale),
conv(args.n_colors * scale * scale, n_feats, kernel_size))
m_body = [
common.OAModule(n_feats,
kernel_size,
use_att=i >= n_resblocks // 4 * 3)
for i in range(n_resblocks)
]
m_body.append(conv(n_feats, 288, kernel_size))
# define tail module
body_tail = [
nn.PixelShuffle(scale),
conv(int(288 // scale // scale), n_feats, kernel_size),
nn.ReLU(True),
conv(n_feats, n_feats, kernel_size),
nn.ReLU(True)
]
m_tail = [
conv(args.n_colors * 3, n_feats, kernel_size),
# common.CBAM(n_feats),
nn.ReLU(True),
conv(n_feats, args.n_colors, kernel_size),
# common.CBAM(n_feats),
# nn.ReLU(False),
# conv(n_feats, args.n_colors, kernel_size)
]
self.head = nn.Sequential(*m_head)
self.body = nn.Sequential(*m_body)
self.body_tail = nn.Sequential(*body_tail)
self.tail = nn.Sequential(*m_tail)
# self.dynamic_kernel0 = conv(n_feats, 64, 7) # common.pixelConv(n_feats, 32, 25, 5, 3)
# self.dynamic_kernel1 = conv(n_feats, 64, 5)
# self.dynamic_kernel2 = conv(n_feats, 64, 3)
self.dynamic_kernel0 = common.pixelConv(n_feats, 64, 25, 5, 3)
self.dynamic_kernel1 = common.pixelConv(n_feats, 64, 49, 7, 3)
self.dynamic_kernel2 = common.pixelConv(n_feats, 64, 9, 3, 3)
self.upsampler = common.bicubic()