def __init__(self, scale_list, model_path = 'weight/EDSR_weight.pt' ):
super(EDSR, self).__init__()
# args
scale = scale_list[0]
input_channel = 5
output_channel = 3
num_block = 16
inp = 64
rgb_range = 255
res_scale = 0.1
act = nn.ReLU(True)
#act = nn.LeakyReLU(negative_slope=0.05, inplace=True)
# head
self.head = nn.Sequential( common.conv(3, inp, input_channel) )
# body
self.body = nn.Sequential( *[ common.ResBlock(inp, bias = True, act = act, res_scale = res_scale) for _ in range( num_block) ] )
self.body.add_module( str(num_block), common.conv(inp, inp, 3) )
# tail
if scale > 1:
self.tail = nn.Sequential( *[ common.Upsampler(scale, inp, act = False, choice = 0),
common.conv(inp, 3, output_channel) ] )
else:
self.tail = nn.Sequential( *[ common.conv(inp, 3, output_channel) ] )
self.sub_mean = common.MeanShift(rgb_range, sign = -1)
self.add_mean = common.MeanShift(rgb_range, sign = 1)
self.model_path = model_path
self.load()
def __init__(self, args):
super().__init__()
n_resgroup = args.n_resgroups
n_feats = args.n_feats
colors = 3
out_dim = 120 # MoL
use_checkpoint = args.checkpoint
self.stages = args.stages
# input = lr + masked_hr + mask [+ vq_latent]
self.head = conv(colors * 2 + 1, n_feats)
self.fuse = conv(n_feats * 2, n_feats, kernel_size=1, bias=False)
self.stage_embedding = nn.Embedding(args.stages, n_feats)
self.body_nl_low = _ResGroup(n_feats,
is_nonlocal=True,
use_checkpoint=use_checkpoint)
self.body = nn.ModuleList(
[_ResGroup(n_feats) for _ in range(n_resgroup - 2)])
self.body_tail = conv(n_feats, n_feats)
self.body_nl_high = _ResGroup(n_feats,
is_nonlocal=True,
use_checkpoint=use_checkpoint)
self.tail = conv(n_feats, out_dim)
if args.position_encoding:
self.position_encoding = nn.Parameter(
torch.zeros(args.stages, n_feats))
else:
self.position_encoding = None
def __init__(self,
inplanes,
planes,
stride=1,
dilation=1,
downsample=None,
conv_type=ConvType.HYPERCUBE,
bn_momentum=0.1,
D=3):
super(BasicBlockBase, self).__init__()
self.conv1 = conv(
inplanes, planes, kernel_size=3, stride=stride, dilation=dilation, conv_type=conv_type, D=D)
self.norm1 = get_norm(self.NORM_TYPE, planes, D, bn_momentum=bn_momentum)
self.conv2 = conv(
planes,
planes,
kernel_size=3,
stride=1,
dilation=dilation,
bias=False,
conv_type=conv_type,
D=D)
self.norm2 = get_norm(self.NORM_TYPE, planes, D, bn_momentum=bn_momentum)
self.relu = MinkowskiReLU(inplace=True)
self.downsample = downsample
def network_initialization(self, in_channels, out_channels, D):
def space_n_time_m(n, m):
return n if D == 3 else [n, n, n, m]
if D == 4:
self.OUT_PIXEL_DIST = space_n_time_m(self.OUT_PIXEL_DIST, 1)
dilations = [1, 1, 1, 1]
bn_momentum = 0.02
self.inplanes = self.INIT_DIM
self.conv1 = conv(in_channels,
self.inplanes,
kernel_size=space_n_time_m(3, 1),
stride=1,
D=D)
self.bn1 = get_norm(NormType.BATCH_NORM,
self.inplanes,
D=self.D,
bn_momentum=bn_momentum)
self.relu = ME.MinkowskiReLU(inplace=True)
self.pool = sum_pool(kernel_size=space_n_time_m(2, 1),
stride=space_n_time_m(2, 1),
D=D)
self.layer1 = self._make_layer(self.BLOCK,
self.PLANES[0],
self.LAYERS[0],
stride=space_n_time_m(2, 1),
dilation=space_n_time_m(
dilations[0], 1))
self.layer2 = self._make_layer(self.BLOCK,
self.PLANES[1],
self.LAYERS[1],
stride=space_n_time_m(2, 1),
dilation=space_n_time_m(
dilations[1], 1))
self.layer3 = self._make_layer(self.BLOCK,
self.PLANES[2],
self.LAYERS[2],
stride=space_n_time_m(2, 1),
dilation=space_n_time_m(
dilations[2], 1))
self.layer4 = self._make_layer(self.BLOCK,
self.PLANES[3],
self.LAYERS[3],
stride=space_n_time_m(2, 1),
dilation=space_n_time_m(
dilations[3], 1))
self.final = conv(self.PLANES[3] * self.BLOCK.expansion,
out_channels,
kernel_size=1,
bias=True,
D=D)
def __init__(self, scale_list, model_path='weight/MDSR_weight.pt'):
super(MDSR, self).__init__()
# args
self.scale_list = scale_list
input_channel = 3
output_channel = 3
num_block = 32
inp = 64
rgb_range = 255
res_scale = 0.1
act = nn.ReLU(True)
#act = nn.LeakyReLU(negative_slope=0.05, inplace=True)
# head
self.head = nn.Sequential(common.conv(3, inp, input_channel))
# pre_process
self.pre_process = nn.ModuleDict([
str(scale),
nn.Sequential(
common.ResBlock(inp, bias=True, act=act, res_scale=res_scale),
common.ResBlock(inp, bias=True, act=act, res_scale=res_scale))
] for scale in self.scale_list)
# body
self.body = nn.Sequential(*[
common.ResBlock(inp, bias=True, act=act, res_scale=res_scale)
for _ in range(num_block)
])
self.body.add_module(str(num_block), common.conv(inp, inp, 3))
#upsample
self.upsample = nn.ModuleDict(
[str(scale),
common.Upsampler(scale, inp, act=False, choice=0)]
for scale in self.scale_list)
# tail
self.tail = nn.Sequential(common.conv(inp, 3, output_channel))
self.sub_mean = common.MeanShift(rgb_range, sign=-1)
self.add_mean = common.MeanShift(rgb_range, sign=1)
self.model_path = model_path
self.load()
def forward(self, inputs):
with tf.variable_scope('backbone_Darknet53'):
# [19, 19, 512], [38, 38, 512], [78, 78, 256]
route_19, route_38, route_76 = self.backbone(
inputs, self.batch_norm_params, self.weight_decay)
with tf.variable_scope('yolov4_head'):
with slim.arg_scope([slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=self.batch_norm_params,
activation_fn=lambda x: tf.nn.leaky_relu(x),
weights_refularizer=slim.l2_regularizer(
self.weight_decay)):
# feature_map_76
fpn_19 = upsample_block(route_19, 256)
route_38 = conv(route_38, 256)
route_38 = tf.concat([route_38, fpn_19], axis=-1)
route_38 = cbl_block(route_38, 256)
fpn_38 = upsample_block(route_38, 128)
route_76 = conv(route_76, 128)
route_76 = tf.concat([route_76, fpn_38], axis=-1)
route_76 = cbl_block(route_76, 128)
pan_76 = route_76
feature_76 = conv(route_76, 256)
# feature_map_38
pan_76 = conv(pan_76, 256, down_sample=True)
route_38 = tf.concat([route_38, pan_76], axis=-1)
route_38 = cbl_block(route_38, 256)
pan_38 = route_38
feature_38 = conv(route_38, 512)
# feature_map_19
pan_38 = conv(pan_38, 512, down_sample=True)
route_19 = tf.concat([route_19, pan_38], axis=-1)
route_19 = cbl_block(route_19, 512)
feature_19 = (route_19, 1024)
feature_19 = slim.conv2d(feature_19,
3 * (4 + 1 + self.class_num),
1,
stride=1,
normalizer_fn=None,
activation_fn=None,
biases_initializer=tf.zeros_initializer())
feature_38 = slim.conv2d(feature_38,
3 * (4 + 1 + self.class_num),
1,
stride=1,
normalizer_fn=None,
activation_fn=None,
biases_initializer=tf.zeros_initializer())
feature_76 = slim.conv2d(feature_76,
3 * (4 + 1 + self.class_num),
1,
stride=1,
normalizer_fn=None,
activation_fn=None,
biases_initializer=tf.zeros_initializer())
return feature_19, feature_38, feature_76
def __init__(self, scale_list, model_path='weight/NEWNET_weight.pt'):
super(NewNet, self).__init__()
# args
self.scale_list = scale_list
self.scale_list.sort()
self.scale_list = list(set(self.scale_list))
input_channel = 3
output_channel = 3
num_block_list = [6,6,6,6]
inp = 64
rgb_range = 255
res_scale = 0.1
act = nn.LeakyReLU(negative_slope=0.05, inplace=True)
# head
self.head = nn.Sequential( common.conv(3, inp, input_channel) )
# body
self.body = nn.ModuleDict([
[ str(scale), nn.Sequential( *[ common.ResBlock(inp, bias = True, act = act, res_scale = res_scale) for _ in range(num_block_list[scale - 1] ) ] ) ] for scale in scale_list
])
#upsample
self.upsample = nn.ModuleDict([
[ str(scale), common.Upsampler(scale, inp, act = False, choice = 0) ] for scale in self.scale_list
])
# tail
self.tail = nn.ModuleDict([
[ str(scale), common.conv(inp, 3, input_channel) ] for scale in self.scale_list
])
self.sub_mean = common.MeanShift(rgb_range, sign = -1)
self.add_mean = common.MeanShift(rgb_range, sign = 1)
self.model_path = model_path
self.load()
def _make_layer(self,
block,
planes,
blocks,
stride=1,
dilation=1,
norm_type=NormType.BATCH_NORM,
bn_momentum=0.1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
conv(self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=stride,
bias=False,
D=self.D),
get_norm(norm_type,
planes * block.expansion,
D=self.D,
bn_momentum=bn_momentum),
)
layers = []
layers.append(
block(self.inplanes,
planes,
stride=stride,
dilation=dilation,
downsample=downsample,
conv_type=self.CONV_TYPE,
D=self.D))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
block(self.inplanes,
planes,
stride=1,
dilation=dilation,
conv_type=self.CONV_TYPE,
D=self.D))
return nn.Sequential(*layers)
def backbone(self, inputs, batch_norm_params, weight_decay):
# =============== Mish activation ===============
with slim.arg_scope(
[slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
activation_fn=Mish, # TODO: Mish func
weights_refularizer=slim.l2_regularizer(weight_decay)):
# CBM
net = conv(inputs, 32)
# CSP1
net = csp_block(net, 32, res_block_sum=1, double_channels=True)
# CSP2
net = csp_block(net, 64, res_block_sum=2, double_channels=False)
# CSP8, feature_map_76
net = csp_block(net, 128, res_block_sum=8, double_channels=False)
route_76 = net
# CSP8, feature_map_38
net = csp_block(net, 256, res_block_sum=8, double_channels=False)
route_38 = net
# CSP4
net = csp_block(net, 512, res_block_sum=4, double_channels=False)
# =============== LeakyRelu activation ===============
with slim.arg_scope(
[slim.conv2d],
normalizer_fn=slim.batch_norm,
normalizer_params=batch_norm_params,
activation_fn=lambda input: tf.nn.leaky_relu(input),
weights_refularizer=slim.l2_regularizer(weight_decay)):
# CBL * 3
net = conv(net, 512, kernel_size=1)
net = conv(net, 1024)
net = conv(net, 512, kernel_size=1)
# SPP, shape:[19, 19, 1024]
net = spp_block(net)
# CBL * 3, feature_map_19
net = conv(net, 512, kernel_size=1)
net = conv(net, 1024)
route_19 = conv(net, 512, kernel_size=1)
return route_19, route_38, route_76
def __init__(self, n_feats, is_nonlocal=False, use_checkpoint=True):
super().__init__()
self.block = ResAttModuleDownUpPlus(n_feats, is_nonlocal,
use_checkpoint)
self.tail = conv(n_feats, n_feats)
def network_initialization(self, in_channels, out_channels, D):
# Setup net_metadata
dilations = self.DILATIONS
bn_momentum = 0.02
def space_n_time_m(n, m):
return n if D == 3 else [n, n, n, m]
if D == 4:
self.OUT_PIXEL_DIST = space_n_time_m(self.OUT_PIXEL_DIST, 1)
# Output of the first conv concated to conv6
self.inplanes = self.INIT_DIM
self.conv0p1s1 = conv(in_channels,
self.inplanes,
kernel_size=space_n_time_m(3, 1),
stride=1,
dilation=1,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bn0 = get_norm(self.NORM_TYPE,
self.inplanes,
D,
bn_momentum=bn_momentum)
self.conv1p1s2 = conv(self.inplanes,
self.inplanes,
kernel_size=space_n_time_m(2, 1),
stride=space_n_time_m(2, 1),
dilation=1,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bn1 = get_norm(self.NORM_TYPE,
self.inplanes,
D,
bn_momentum=bn_momentum)
self.block1 = self._make_layer(self.BLOCK,
self.PLANES[0],
self.LAYERS[0],
dilation=dilations[0],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.conv2p2s2 = conv(self.inplanes,
self.inplanes,
kernel_size=space_n_time_m(2, 1),
stride=space_n_time_m(2, 1),
dilation=1,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bn2 = get_norm(self.NORM_TYPE,
self.inplanes,
D,
bn_momentum=bn_momentum)
self.block2 = self._make_layer(self.BLOCK,
self.PLANES[1],
self.LAYERS[1],
dilation=dilations[1],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.conv3p4s2 = conv(self.inplanes,
self.inplanes,
kernel_size=space_n_time_m(2, 1),
stride=space_n_time_m(2, 1),
dilation=1,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bn3 = get_norm(self.NORM_TYPE,
self.inplanes,
D,
bn_momentum=bn_momentum)
self.block3 = self._make_layer(self.BLOCK,
self.PLANES[2],
self.LAYERS[2],
dilation=dilations[2],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.conv4p8s2 = conv(self.inplanes,
self.inplanes,
kernel_size=space_n_time_m(2, 1),
stride=space_n_time_m(2, 1),
dilation=1,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bn4 = get_norm(self.NORM_TYPE,
self.inplanes,
D,
bn_momentum=bn_momentum)
self.block4 = self._make_layer(self.BLOCK,
self.PLANES[3],
self.LAYERS[3],
dilation=dilations[3],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.convtr4p16s2 = conv_tr(self.inplanes,
self.PLANES[4],
kernel_size=space_n_time_m(2, 1),
upsample_stride=space_n_time_m(2, 1),
dilation=1,
bias=False,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bntr4 = get_norm(self.NORM_TYPE,
self.PLANES[4],
D,
bn_momentum=bn_momentum)
self.inplanes = self.PLANES[4] + self.PLANES[2] * self.BLOCK.expansion
self.block5 = self._make_layer(self.BLOCK,
self.PLANES[4],
self.LAYERS[4],
dilation=dilations[4],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.convtr5p8s2 = conv_tr(self.inplanes,
self.PLANES[5],
kernel_size=space_n_time_m(2, 1),
upsample_stride=space_n_time_m(2, 1),
dilation=1,
bias=False,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bntr5 = get_norm(self.NORM_TYPE,
self.PLANES[5],
D,
bn_momentum=bn_momentum)
self.inplanes = self.PLANES[5] + self.PLANES[1] * self.BLOCK.expansion
self.block6 = self._make_layer(self.BLOCK,
self.PLANES[5],
self.LAYERS[5],
dilation=dilations[5],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.convtr6p4s2 = conv_tr(self.inplanes,
self.PLANES[6],
kernel_size=space_n_time_m(2, 1),
upsample_stride=space_n_time_m(2, 1),
dilation=1,
bias=False,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bntr6 = get_norm(self.NORM_TYPE,
self.PLANES[6],
D,
bn_momentum=bn_momentum)
self.inplanes = self.PLANES[6] + self.PLANES[0] * self.BLOCK.expansion
self.block7 = self._make_layer(self.BLOCK,
self.PLANES[6],
self.LAYERS[6],
dilation=dilations[6],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.convtr7p2s2 = conv_tr(self.inplanes,
self.PLANES[7],
kernel_size=space_n_time_m(2, 1),
upsample_stride=space_n_time_m(2, 1),
dilation=1,
bias=False,
conv_type=self.NON_BLOCK_CONV_TYPE,
D=D)
self.bntr7 = get_norm(self.NORM_TYPE,
self.PLANES[7],
D,
bn_momentum=bn_momentum)
self.inplanes = self.PLANES[7] + self.INIT_DIM
self.block8 = self._make_layer(self.BLOCK,
self.PLANES[7],
self.LAYERS[7],
dilation=dilations[7],
norm_type=self.NORM_TYPE,
bn_momentum=bn_momentum)
self.final = conv(self.PLANES[7],
out_channels,
kernel_size=1,
stride=1,
bias=True,
D=D)
self.relu = MinkowskiReLU(inplace=True)