vismatch 1.1.1__py3-none-any.whl

vismatch/third_party/MatchAnything/imcui/third_party/MatchAnything/src/loftr/backbone/resnet_fpn.py

@@ -0,0 +1,1094 @@
+import torch.nn as nn
+import torch.nn.functional as F
+from .repvgg import get_RepVGG_func_by_name
+from .s2dnet import S2DNet
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    """1x1 convolution without padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, padding=0, bias=False)
+
+
+def conv3x3(in_planes, out_planes, stride=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)
+
+
+class BasicBlock(nn.Module):
+    def __init__(self, in_planes, planes, stride=1):
+        super().__init__()
+        self.conv1 = conv3x3(in_planes, planes, stride)
+        self.conv2 = conv3x3(planes, planes)
+        self.bn1 = nn.BatchNorm2d(planes)
+        self.bn2 = nn.BatchNorm2d(planes)
+        self.relu = nn.ReLU(inplace=True)
+
+        if stride == 1:
+            self.downsample = None
+        else:
+            self.downsample = nn.Sequential(
+                conv1x1(in_planes, planes, stride=stride),
+                nn.BatchNorm2d(planes)
+            )
+
+    def forward(self, x):
+        y = x
+        y = self.relu(self.bn1(self.conv1(y)))
+        y = self.bn2(self.conv2(y))
+
+        if self.downsample is not None:
+            x = self.downsample(x)
+
+        return self.relu(x+y)
+
+
+class ResNetFPN_8_2(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1/2.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+        self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+        self.layer2_outconv2 = nn.Sequential(
+            conv3x3(block_dims[2], block_dims[2]),
+            nn.BatchNorm2d(block_dims[2]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[2], block_dims[1]),
+        )
+        self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+        self.layer1_outconv2 = nn.Sequential(
+            conv3x3(block_dims[1], block_dims[1]),
+            nn.BatchNorm2d(block_dims[1]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[1], block_dims[0]),
+        )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        # FPN
+        x3_out = self.layer3_outconv(x3)
+
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        return {'feats_c': x3_out, 'feats_f': x1_out}
+    
+    def pro(self, x, profiler):
+        with profiler.profile('ResNet Backbone'):
+            # ResNet Backbone
+            x0 = self.relu(self.bn1(self.conv1(x)))
+            x1 = self.layer1(x0)  # 1/2
+            x2 = self.layer2(x1)  # 1/4
+            x3 = self.layer3(x2)  # 1/8
+
+        with profiler.profile('ResNet FPN'):
+            # FPN
+            x3_out = self.layer3_outconv(x3)
+
+            x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=True)
+            x2_out = self.layer2_outconv(x2)
+            x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+            x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=True)
+            x1_out = self.layer1_outconv(x1)
+            x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+            return {'feats_c': x3_out, 'feats_f': x1_out}
+
+class ResNetFPN_8_2_fix(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1/2.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+        self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+        self.layer2_outconv2 = nn.Sequential(
+            conv3x3(block_dims[2], block_dims[2]),
+            nn.BatchNorm2d(block_dims[2]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[2], block_dims[1]),
+        )
+        self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+        self.layer1_outconv2 = nn.Sequential(
+            conv3x3(block_dims[1], block_dims[1]),
+            nn.BatchNorm2d(block_dims[1]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[1], block_dims[0]),
+        )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        # FPN
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else:
+                return {'feats_c': x3, 'feats_f': None}
+
+
+        x3_out = self.layer3_outconv(x3) # n+1
+
+        x3_out_2x = F.interpolate(x3_out, size=((x3_out.size(-2)-1)*2+1, (x3_out.size(-1)-1)*2+1), mode='bilinear', align_corners=True) # 2n+1
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, size=((x2_out.size(-2)-1)*2+1, (x2_out.size(-1)-1)*2+1), mode='bilinear', align_corners=True) # 4n+1
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        return {'feats_c': x3_out, 'feats_f': x1_out}
+
+
+class ResNetFPN_16_4(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/16 and 1/4.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+        self.layer4 = self._make_layer(block, block_dims[3], stride=2)  # 1/16
+
+        # 3. FPN upsample
+        self.layer4_outconv = conv1x1(block_dims[3], block_dims[3])
+        self.layer3_outconv = conv1x1(block_dims[2], block_dims[3])
+        self.layer3_outconv2 = nn.Sequential(
+            conv3x3(block_dims[3], block_dims[3]),
+            nn.BatchNorm2d(block_dims[3]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[3], block_dims[2]),
+        )
+
+        self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+        self.layer2_outconv2 = nn.Sequential(
+            conv3x3(block_dims[2], block_dims[2]),
+            nn.BatchNorm2d(block_dims[2]),
+            nn.LeakyReLU(),
+            conv3x3(block_dims[2], block_dims[1]),
+        )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+        x4 = self.layer4(x3)  # 1/16
+
+        # FPN
+        x4_out = self.layer4_outconv(x4)
+
+        x4_out_2x = F.interpolate(x4_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x3_out = self.layer3_outconv(x3)
+        x3_out = self.layer3_outconv2(x3_out+x4_out_2x)
+
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        return {'feats_c': x4_out, 'feats_f': x2_out}
+
+
+class ResNetFPN_8_1(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+            self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+            self.layer2_outconv2 = nn.Sequential(
+                conv3x3(block_dims[2], block_dims[2]),
+                nn.BatchNorm2d(block_dims[2]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[2], block_dims[1]),
+            )
+            self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+            self.layer1_outconv2 = nn.Sequential(
+                conv3x3(block_dims[1], block_dims[1]),
+                nn.BatchNorm2d(block_dims[1]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[1], block_dims[0]),
+            )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        # FPN
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else:
+                return {'feats_c': x3, 'feats_f': None}
+
+        x3_out = self.layer3_outconv(x3)
+
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=True)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        if not self.inter_feat:
+            return {'feats_c': x3, 'feats_f': x0_out}
+        else:
+            return {'feats_c': x3, 'feats_f': x0_out, 'feats_x2': x2, 'feats_x1': x1}
+
+
+class ResNetFPN_8_1_align(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+            self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+            self.layer2_outconv2 = nn.Sequential(
+                conv3x3(block_dims[2], block_dims[2]),
+                nn.BatchNorm2d(block_dims[2]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[2], block_dims[1]),
+            )
+            self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+            self.layer1_outconv2 = nn.Sequential(
+                conv3x3(block_dims[1], block_dims[1]),
+                nn.BatchNorm2d(block_dims[1]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[1], block_dims[0]),
+            )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        # FPN
+        
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else:
+                return {'feats_c': x3, 'feats_f': None}
+                
+        x3_out = self.layer3_outconv(x3)
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        if not self.inter_feat:
+            return {'feats_c': x3, 'feats_f': x0_out}
+        else:
+            return {'feats_c': x3, 'feats_f': x0_out, 'feats_x2': x2, 'feats_x1': x1}
+    
+    def pro(self, x, profiler):
+        with profiler.profile('ResNet Backbone'):
+            # ResNet Backbone
+            x0 = self.relu(self.bn1(self.conv1(x)))
+            x1 = self.layer1(x0)  # 1/2
+            x2 = self.layer2(x1)  # 1/4
+            x3 = self.layer3(x2)  # 1/8
+
+        with profiler.profile('FPN'):
+            # FPN
+            x3_out = self.layer3_outconv(x3)
+            
+            if self.skip_fine:
+                return [x3_out, None]
+            x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+            x2_out = self.layer2_outconv(x2)
+            x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+            x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+            x1_out = self.layer1_outconv(x1)
+            x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        with profiler.profile('upsample*1'):
+            x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        return {'feats_c': x3_out, 'feats_f': x0_out}
+    
+
+class ResNetFPN_8_2_align(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1/2.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+            self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+            self.layer2_outconv2 = nn.Sequential(
+                conv3x3(block_dims[2], block_dims[2]),
+                nn.BatchNorm2d(block_dims[2]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[2], block_dims[1]),
+            )
+            self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+            self.layer1_outconv2 = nn.Sequential(
+                conv3x3(block_dims[1], block_dims[1]),
+                nn.BatchNorm2d(block_dims[1]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[1], block_dims[0]),
+            )
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else:
+                return {'feats_c': x3, 'feats_f': None}
+
+        # FPN
+        x3_out = self.layer3_outconv(x3)
+
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        if not self.inter_feat:
+            return {'feats_c': x3, 'feats_f': x1_out}
+        else:
+            return {'feats_c': x3, 'feats_f': x1_out, 'feats_x2': x2, 'feats_x1': x1}
+
+
+class ResNet_8_1_align(nn.Module):
+    """
+    ResNet, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+
+        # Networks
+        self.conv1 = nn.Conv2d(1, initial_dim, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = nn.BatchNorm2d(initial_dim)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.layer1 = self._make_layer(block, block_dims[0], stride=1)  # 1/2
+        self.layer2 = self._make_layer(block, block_dims[1], stride=2)  # 1/4
+        self.layer3 = self._make_layer(block, block_dims[2], stride=2)  # 1/8
+
+        # 3. FPN upsample
+        # self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+        # self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+        # self.layer2_outconv2 = nn.Sequential(
+        #     conv3x3(block_dims[2], block_dims[2]),
+        #     nn.BatchNorm2d(block_dims[2]),
+        #     nn.LeakyReLU(),
+        #     conv3x3(block_dims[2], block_dims[1]),
+        # )
+        # self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+        # self.layer1_outconv2 = nn.Sequential(
+        #     conv3x3(block_dims[1], block_dims[1]),
+        #     nn.BatchNorm2d(block_dims[1]),
+        #     nn.LeakyReLU(),
+        #     conv3x3(block_dims[1], block_dims[0]),
+        # )
+        self.layer0_outconv = conv1x1(block_dims[2], block_dims[0])
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # ResNet Backbone
+        x0 = self.relu(self.bn1(self.conv1(x)))
+        x1 = self.layer1(x0)  # 1/2
+        x2 = self.layer2(x1)  # 1/4
+        x3 = self.layer3(x2)  # 1/8
+
+        # FPN
+        # x3_out = self.layer3_outconv(x3)
+
+        # x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x2_out = self.layer2_outconv(x2)
+        # x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        # x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x1_out = self.layer1_outconv(x1)
+        # x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        x0_out = F.interpolate(x3, scale_factor=8., mode='bilinear', align_corners=False)
+        x0_out = self.layer0_outconv(x0_out)
+
+        return {'feats_c': x3, 'feats_f': x0_out}
+
+class VGG_8_1_align(nn.Module):
+    """
+    VGG-like backbone, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        self.relu = nn.ReLU(inplace=True)
+        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
+        c1, c2, c3, c4, c5 = 64, 64, 128, 128, 256
+
+        self.conv1a = nn.Conv2d(1, c1, kernel_size=3, stride=1, padding=1)
+        self.conv1b = nn.Conv2d(c1, c1, kernel_size=3, stride=1, padding=1)
+        self.conv2a = nn.Conv2d(c1, c2, kernel_size=3, stride=1, padding=1)
+        self.conv2b = nn.Conv2d(c2, c2, kernel_size=3, stride=1, padding=1)
+        self.conv3a = nn.Conv2d(c2, c3, kernel_size=3, stride=1, padding=1)
+        self.conv3b = nn.Conv2d(c3, c3, kernel_size=3, stride=1, padding=1)
+        self.conv4a = nn.Conv2d(c3, c4, kernel_size=3, stride=1, padding=1)
+        self.conv4b = nn.Conv2d(c4, c4, kernel_size=3, stride=1, padding=1)
+
+        # self.convPa = nn.Conv2d(c4, c5, kernel_size=3, stride=1, padding=1)
+        # self.convPb = nn.Conv2d(c5, 65, kernel_size=1, stride=1, padding=0)
+
+        self.convDa = nn.Conv2d(c4, c5, kernel_size=3, stride=1, padding=1)
+        self.convDb = nn.Conv2d(
+            c5, 256,
+            kernel_size=1, stride=1, padding=0)
+        self.layer0_outconv = conv1x1(block_dims[2], block_dims[0])
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, dim, stride=1):
+        layer1 = block(self.in_planes, dim, stride=stride)
+        layer2 = block(dim, dim, stride=1)
+        layers = (layer1, layer2)
+
+        self.in_planes = dim
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        # Shared Encoder
+        x = self.relu(self.conv1a(x))
+        x = self.relu(self.conv1b(x))
+        x = self.pool(x)
+        x = self.relu(self.conv2a(x))
+        x = self.relu(self.conv2b(x))
+        x = self.pool(x)
+        x = self.relu(self.conv3a(x))
+        x = self.relu(self.conv3b(x))
+        x = self.pool(x)
+        x = self.relu(self.conv4a(x))
+        x = self.relu(self.conv4b(x))
+
+        cDa = self.relu(self.convDa(x))
+        descriptors = self.convDb(cDa)
+        x3_out = nn.functional.normalize(descriptors, p=2, dim=1)
+
+        x0_out = F.interpolate(x3_out, scale_factor=8., mode='bilinear', align_corners=False)
+        x0_out = self.layer0_outconv(x0_out)
+        # ResNet Backbone
+        # x0 = self.relu(self.bn1(self.conv1(x)))
+        # x1 = self.layer1(x0)  # 1/2
+        # x2 = self.layer2(x1)  # 1/4
+        # x3 = self.layer3(x2)  # 1/8
+
+        # # FPN
+        # x3_out = self.layer3_outconv(x3)
+
+        # x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x2_out = self.layer2_outconv(x2)
+        # x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        # x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x1_out = self.layer1_outconv(x1)
+        # x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        # x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        return {'feats_c': x3_out, 'feats_f': x0_out}
+
+class RepVGG_8_1_align(nn.Module):
+    """
+    RepVGG backbone, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        # block = BasicBlock
+        # initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+        self.leaky = config['leaky']
+
+        # backbone_name='RepVGG-B0'
+        if config.get('repvggmodel') is not None:
+            backbone_name=config['repvggmodel']
+        elif self.leaky:
+            backbone_name='RepVGG-A1_leaky'
+        else:
+            backbone_name='RepVGG-A1'
+        repvgg_fn = get_RepVGG_func_by_name(backbone_name)
+        backbone = repvgg_fn(False)
+        self.layer0, self.layer1, self.layer2, self.layer3 = backbone.stage0, backbone.stage1, backbone.stage2, backbone.stage3 #, backbone.stage4
+        # self.layer0, self.layer1, self.layer2, self.layer3, self.layer4 = backbone.stage0, backbone.stage1, backbone.stage2, backbone.stage3, backbone.stage4
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+            self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+            self.layer2_outconv2 = nn.Sequential(
+                conv3x3(block_dims[2], block_dims[2]),
+                nn.BatchNorm2d(block_dims[2]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[2], block_dims[1]),
+            )
+            self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+            self.layer1_outconv2 = nn.Sequential(
+                conv3x3(block_dims[1], block_dims[1]),
+                nn.BatchNorm2d(block_dims[1]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[1], block_dims[0]),
+            )
+
+        # self.layer0_outconv = conv1x1(192, 48)
+
+        for layer in [self.layer0, self.layer1, self.layer2, self.layer3]:
+            for m in layer.modules():
+                if isinstance(m, nn.Conv2d):
+                    nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                    nn.init.constant_(m.weight, 1)
+                    nn.init.constant_(m.bias, 0)
+        # for layer in [self.layer0, self.layer1, self.layer2, self.layer3, self.layer4]:
+        #     for m in layer.modules():
+        #         if isinstance(m, nn.Conv2d):
+        #             nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+        #         elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+        #             nn.init.constant_(m.weight, 1)
+        #             nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+
+        out = self.layer0(x) # 1/2
+        for module in self.layer1:
+            out = module(out) # 1/2
+            x1 = out
+        for module in self.layer2:
+            out = module(out) # 1/4
+            x2 = out
+        for module in self.layer3:
+            out = module(out) # 1/8
+            x3 = out
+        # for module in self.layer4:
+        #     out = module(out)
+        #     x3 = out
+                
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else: 
+                return {'feats_c': x3, 'feats_f': None}
+        x3_out = self.layer3_outconv(x3)
+        x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        # x_f = F.interpolate(x_c, scale_factor=8., mode='bilinear', align_corners=False)
+        # x_f = self.layer0_outconv(x_f)
+        return {'feats_c': x3_out, 'feats_f': x0_out}
+
+
+class RepVGG_8_2_fix(nn.Module):
+    """
+    RepVGG backbone, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        # block = BasicBlock
+        # initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+
+        # backbone_name='RepVGG-B0'
+        backbone_name='RepVGG-A1'
+        repvgg_fn = get_RepVGG_func_by_name(backbone_name)
+        backbone = repvgg_fn(False)
+        self.layer0, self.layer1, self.layer2, self.layer3 = backbone.stage0, backbone.stage1, backbone.stage2, backbone.stage3 #, backbone.stage4
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+            self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+            self.layer2_outconv2 = nn.Sequential(
+                conv3x3(block_dims[2], block_dims[2]),
+                nn.BatchNorm2d(block_dims[2]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[2], block_dims[1]),
+            )
+            self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+            self.layer1_outconv2 = nn.Sequential(
+                conv3x3(block_dims[1], block_dims[1]),
+                nn.BatchNorm2d(block_dims[1]),
+                nn.LeakyReLU(),
+                conv3x3(block_dims[1], block_dims[0]),
+            )
+
+        # self.layer0_outconv = conv1x1(192, 48)
+
+        for layer in [self.layer0, self.layer1, self.layer2, self.layer3]:
+            for m in layer.modules():
+                if isinstance(m, nn.Conv2d):
+                    nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                    nn.init.constant_(m.weight, 1)
+                    nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+
+        x0 = self.layer0(x) # 1/2
+        out = x0
+        for module in self.layer1:
+            out = module(out) # 1/2
+            x1 = out
+        for module in self.layer2:
+            out = module(out) # 1/4
+            x2 = out
+        for module in self.layer3:
+            out = module(out) # 1/8
+            x3 = out
+        # for module in self.layer4:
+        #     out = module(out)
+        
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else: 
+                return {'feats_c': x3, 'feats_f': None}
+        x3_out = self.layer3_outconv(x3)
+        x3_out_2x = F.interpolate(x3_out, size=((x3_out.size(-2)-1)*2+1, (x3_out.size(-1)-1)*2+1), mode='bilinear', align_corners=True)
+        x2_out = self.layer2_outconv(x2)
+        x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        x2_out_2x = F.interpolate(x2_out, size=((x2_out.size(-2)-1)*2+1, (x2_out.size(-1)-1)*2+1), mode='bilinear', align_corners=True)
+        x1_out = self.layer1_outconv(x1)
+        x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        # x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        # x_f = F.interpolate(x_c, scale_factor=8., mode='bilinear', align_corners=False)
+        # x_f = self.layer0_outconv(x_f)
+        return {'feats_c': x3_out, 'feats_f': x1_out}
+
+
+class RepVGGnfpn_8_1_align(nn.Module):
+    """
+    RepVGG backbone, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        # block = BasicBlock
+        # initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+
+        # backbone_name='RepVGG-B0'
+        backbone_name='RepVGG-A1'
+        repvgg_fn = get_RepVGG_func_by_name(backbone_name)
+        backbone = repvgg_fn(False)
+        self.layer0, self.layer1, self.layer2, self.layer3 = backbone.stage0, backbone.stage1, backbone.stage2, backbone.stage3 #, backbone.stage4
+
+        # 3. FPN upsample
+        if not self.skip_fine_feature:
+            self.layer0_outconv = conv1x1(block_dims[2], block_dims[0])
+        # self.layer0_outconv = conv1x1(192, 48)
+
+        for layer in [self.layer0, self.layer1, self.layer2, self.layer3]:
+            for m in layer.modules():
+                if isinstance(m, nn.Conv2d):
+                    nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                    nn.init.constant_(m.weight, 1)
+                    nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+
+        x0 = self.layer0(x) # 1/2
+        out = x0
+        for module in self.layer1:
+            out = module(out) # 1/2
+            x1 = out
+        for module in self.layer2:
+            out = module(out) # 1/4
+            x2 = out
+        for module in self.layer3:
+            out = module(out) # 1/8
+            x3 = out
+        # for module in self.layer4:
+        #     out = module(out)
+        
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': x3, 'feats_f': None, 'feats_x2': x2, 'feats_x1': x1}
+            else: 
+                return {'feats_c': x3, 'feats_f': None}
+        # x3_out = self.layer3_outconv(x3)
+        # x3_out_2x = F.interpolate(x3_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x2_out = self.layer2_outconv(x2)
+        # x2_out = self.layer2_outconv2(x2_out+x3_out_2x)
+
+        # x2_out_2x = F.interpolate(x2_out, scale_factor=2., mode='bilinear', align_corners=False)
+        # x1_out = self.layer1_outconv(x1)
+        # x1_out = self.layer1_outconv2(x1_out+x2_out_2x)
+
+        # x0_out = F.interpolate(x1_out, scale_factor=2., mode='bilinear', align_corners=False)
+
+        x_f = F.interpolate(x3, scale_factor=8., mode='bilinear', align_corners=False)
+        x_f = self.layer0_outconv(x_f)
+        # x_f2 = F.interpolate(x3, scale_factor=8., mode='bilinear', align_corners=False)
+        # x_f2 = self.layer0_outconv(x_f2)
+        return {'feats_c': x3, 'feats_f': x_f}
+
+
+class s2dnet_8_1_align(nn.Module):
+    """
+    ResNet+FPN, output resolution are 1/8 and 1.
+    Each block has 2 layers.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        # Config
+        block = BasicBlock
+        initial_dim = config['initial_dim']
+        block_dims = config['block_dims']
+
+        # Class Variable
+        self.in_planes = initial_dim
+        self.skip_fine_feature = config['coarse_feat_only']
+        self.inter_feat = config['inter_feat']
+        # Networks
+        self.backbone = S2DNet(checkpoint_path = '/cephfs-mvs/3dv-research/hexingyi/code_yf/loftrdev/weights/s2dnet/s2dnet_weights.pth')
+        # 3. FPN upsample
+        # self.layer3_outconv = conv1x1(block_dims[2], block_dims[2])
+        # if not self.skip_fine_feature:
+        #     self.layer2_outconv = conv1x1(block_dims[1], block_dims[2])
+        #     self.layer2_outconv2 = nn.Sequential(
+        #         conv3x3(block_dims[2], block_dims[2]),
+        #         nn.BatchNorm2d(block_dims[2]),
+        #         nn.LeakyReLU(),
+        #         conv3x3(block_dims[2], block_dims[1]),
+        #     )
+        #     self.layer1_outconv = conv1x1(block_dims[0], block_dims[1])
+        #     self.layer1_outconv2 = nn.Sequential(
+        #         conv3x3(block_dims[1], block_dims[1]),
+        #         nn.BatchNorm2d(block_dims[1]),
+        #         nn.LeakyReLU(),
+        #         conv3x3(block_dims[1], block_dims[0]),
+        #     )
+
+        # for m in self.modules():
+        #     if isinstance(m, nn.Conv2d):
+        #         nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+        #     elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+        #         nn.init.constant_(m.weight, 1)
+        #         nn.init.constant_(m.bias, 0)
+
+    def forward(self, x):
+        ret = self.backbone(x)
+        ret[2] = F.interpolate(ret[2], scale_factor=2., mode='bilinear', align_corners=False)
+        if self.skip_fine_feature:
+            if self.inter_feat:
+                return {'feats_c': ret[2], 'feats_f': None, 'feats_x2': ret[1], 'feats_x1': ret[0]}
+            else:
+                return {'feats_c': ret[2], 'feats_f': None,}
+    
+    def pro(self, x, profiler):
+        pass