vismatch 1.1.1__py3-none-any.whl

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/backbone/__init__.py

@@ -0,0 +1,11 @@
+from .resnet_fpn import ResNetFPN_8_2, ResNetFPN_16_4
+
+
+def build_backbone(config):
+    if config['backbone_type'] == 'ResNetFPN':
+        if config['resolution'] == (8, 2):
+            return ResNetFPN_8_2(config['resnetfpn'])
+        elif config['resolution'] == (16, 4):
+            return ResNetFPN_16_4(config['resnetfpn'])
+    else:
+        raise ValueError(f"LOFTR.BACKBONE_TYPE {config['backbone_type']} not supported.")

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/backbone/resnet_fpn.py

@@ -0,0 +1,199 @@
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+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 [x3_out, 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 [x4_out, x2_out]

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/loftr.py

@@ -0,0 +1,81 @@
+import torch
+import torch.nn as nn
+from einops.einops import rearrange
+
+from .backbone import build_backbone
+from .utils.position_encoding import PositionEncodingSine
+from .loftr_module import LocalFeatureTransformer, FinePreprocess
+from .utils.coarse_matching import CoarseMatching
+from .utils.fine_matching import FineMatching
+
+
+class LoFTR(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        # Misc
+        self.config = config
+
+        # Modules
+        self.backbone = build_backbone(config)
+        self.pos_encoding = PositionEncodingSine(
+            config['coarse']['d_model'],
+            temp_bug_fix=config['coarse']['temp_bug_fix'])
+        self.loftr_coarse = LocalFeatureTransformer(config['coarse'])
+        self.coarse_matching = CoarseMatching(config['match_coarse'])
+        self.fine_preprocess = FinePreprocess(config)
+        self.loftr_fine = LocalFeatureTransformer(config["fine"])
+        self.fine_matching = FineMatching()
+
+    def forward(self, data):
+        """ 
+        Update:
+            data (dict): {
+                'image0': (torch.Tensor): (N, 1, H, W)
+                'image1': (torch.Tensor): (N, 1, H, W)
+                'mask0'(optional) : (torch.Tensor): (N, H, W) '0' indicates a padded position
+                'mask1'(optional) : (torch.Tensor): (N, H, W)
+            }
+        """
+        # 1. Local Feature CNN
+        data.update({
+            'bs': data['image0'].size(0),
+            'hw0_i': data['image0'].shape[2:], 'hw1_i': data['image1'].shape[2:]
+        })
+
+        if data['hw0_i'] == data['hw1_i']:  # faster & better BN convergence
+            feats_c, feats_f = self.backbone(torch.cat([data['image0'], data['image1']], dim=0))
+            (feat_c0, feat_c1), (feat_f0, feat_f1) = feats_c.split(data['bs']), feats_f.split(data['bs'])
+        else:  # handle different input shapes
+            (feat_c0, feat_f0), (feat_c1, feat_f1) = self.backbone(data['image0']), self.backbone(data['image1'])
+
+        data.update({
+            'hw0_c': feat_c0.shape[2:], 'hw1_c': feat_c1.shape[2:],
+            'hw0_f': feat_f0.shape[2:], 'hw1_f': feat_f1.shape[2:]
+        })
+
+        # 2. coarse-level loftr module
+        # add featmap with positional encoding, then flatten it to sequence [N, HW, C]
+        feat_c0 = rearrange(self.pos_encoding(feat_c0), 'n c h w -> n (h w) c')
+        feat_c1 = rearrange(self.pos_encoding(feat_c1), 'n c h w -> n (h w) c')
+
+        mask_c0 = mask_c1 = None  # mask is useful in training
+        if 'mask0' in data:
+            mask_c0, mask_c1 = data['mask0'].flatten(-2), data['mask1'].flatten(-2)
+        feat_c0, feat_c1 = self.loftr_coarse(feat_c0, feat_c1, mask_c0, mask_c1)
+
+        # 3. match coarse-level
+        self.coarse_matching(feat_c0, feat_c1, data, mask_c0=mask_c0, mask_c1=mask_c1)
+
+        # 4. fine-level refinement
+        feat_f0_unfold, feat_f1_unfold = self.fine_preprocess(feat_f0, feat_f1, feat_c0, feat_c1, data)
+        if feat_f0_unfold.size(0) != 0:  # at least one coarse level predicted
+            feat_f0_unfold, feat_f1_unfold = self.loftr_fine(feat_f0_unfold, feat_f1_unfold)
+
+        # 5. match fine-level
+        self.fine_matching(feat_f0_unfold, feat_f1_unfold, data)
+
+    def load_state_dict(self, state_dict, *args, **kwargs):
+        for k in list(state_dict.keys()):
+            if k.startswith('matcher.'):
+                state_dict[k.replace('matcher.', '', 1)] = state_dict.pop(k)
+        return super().load_state_dict(state_dict, *args, **kwargs)

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/loftr_module/__init__.py

@@ -0,0 +1,2 @@
+from .transformer import LocalFeatureTransformer
+from .fine_preprocess import FinePreprocess

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/loftr_module/fine_preprocess.py

@@ -0,0 +1,59 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from einops.einops import rearrange, repeat
+
+
+class FinePreprocess(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+        self.cat_c_feat = config['fine_concat_coarse_feat']
+        self.W = self.config['fine_window_size']
+
+        d_model_c = self.config['coarse']['d_model']
+        d_model_f = self.config['fine']['d_model']
+        self.d_model_f = d_model_f
+        if self.cat_c_feat:
+            self.down_proj = nn.Linear(d_model_c, d_model_f, bias=True)
+            self.merge_feat = nn.Linear(2*d_model_f, d_model_f, bias=True)
+
+        self._reset_parameters()
+
+    def _reset_parameters(self):
+        for p in self.parameters():
+            if p.dim() > 1:
+                nn.init.kaiming_normal_(p, mode="fan_out", nonlinearity="relu")
+
+    def forward(self, feat_f0, feat_f1, feat_c0, feat_c1, data):
+        W = self.W
+        stride = data['hw0_f'][0] // data['hw0_c'][0]
+
+        data.update({'W': W})
+        if data['b_ids'].shape[0] == 0:
+            feat0 = torch.empty(0, self.W**2, self.d_model_f, device=feat_f0.device)
+            feat1 = torch.empty(0, self.W**2, self.d_model_f, device=feat_f0.device)
+            return feat0, feat1
+
+        # 1. unfold(crop) all local windows
+        feat_f0_unfold = F.unfold(feat_f0, kernel_size=(W, W), stride=stride, padding=W//2)
+        feat_f0_unfold = rearrange(feat_f0_unfold, 'n (c ww) l -> n l ww c', ww=W**2)
+        feat_f1_unfold = F.unfold(feat_f1, kernel_size=(W, W), stride=stride, padding=W//2)
+        feat_f1_unfold = rearrange(feat_f1_unfold, 'n (c ww) l -> n l ww c', ww=W**2)
+
+        # 2. select only the predicted matches
+        feat_f0_unfold = feat_f0_unfold[data['b_ids'], data['i_ids']]  # [n, ww, cf]
+        feat_f1_unfold = feat_f1_unfold[data['b_ids'], data['j_ids']]
+
+        # option: use coarse-level loftr feature as context: concat and linear
+        if self.cat_c_feat:
+            feat_c_win = self.down_proj(torch.cat([feat_c0[data['b_ids'], data['i_ids']],
+                                                   feat_c1[data['b_ids'], data['j_ids']]], 0))  # [2n, c]
+            feat_cf_win = self.merge_feat(torch.cat([
+                torch.cat([feat_f0_unfold, feat_f1_unfold], 0),  # [2n, ww, cf]
+                repeat(feat_c_win, 'n c -> n ww c', ww=W**2),  # [2n, ww, cf]
+            ], -1))
+            feat_f0_unfold, feat_f1_unfold = torch.chunk(feat_cf_win, 2, dim=0)
+
+        return feat_f0_unfold, feat_f1_unfold

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/loftr_module/linear_attention.py

@@ -0,0 +1,81 @@
+"""
+Linear Transformer proposed in "Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention"
+Modified from: https://github.com/idiap/fast-transformers/blob/master/fast_transformers/attention/linear_attention.py
+"""
+
+import torch
+from torch.nn import Module, Dropout
+
+
+def elu_feature_map(x):
+    return torch.nn.functional.elu(x) + 1
+
+
+class LinearAttention(Module):
+    def __init__(self, eps=1e-6):
+        super().__init__()
+        self.feature_map = elu_feature_map
+        self.eps = eps
+
+    def forward(self, queries, keys, values, q_mask=None, kv_mask=None):
+        """ Multi-Head linear attention proposed in "Transformers are RNNs"
+        Args:
+            queries: [N, L, H, D]
+            keys: [N, S, H, D]
+            values: [N, S, H, D]
+            q_mask: [N, L]
+            kv_mask: [N, S]
+        Returns:
+            queried_values: (N, L, H, D)
+        """
+        Q = self.feature_map(queries)
+        K = self.feature_map(keys)
+
+        # set padded position to zero
+        if q_mask is not None:
+            Q = Q * q_mask[:, :, None, None]
+        if kv_mask is not None:
+            K = K * kv_mask[:, :, None, None]
+            values = values * kv_mask[:, :, None, None]
+
+        v_length = values.size(1)
+        values = values / v_length  # prevent fp16 overflow
+        KV = torch.einsum("nshd,nshv->nhdv", K, values)  # (S,D)' @ S,V
+        Z = 1 / (torch.einsum("nlhd,nhd->nlh", Q, K.sum(dim=1)) + self.eps)
+        queried_values = torch.einsum("nlhd,nhdv,nlh->nlhv", Q, KV, Z) * v_length
+
+        return queried_values.contiguous()
+
+
+class FullAttention(Module):
+    def __init__(self, use_dropout=False, attention_dropout=0.1):
+        super().__init__()
+        self.use_dropout = use_dropout
+        self.dropout = Dropout(attention_dropout)
+
+    def forward(self, queries, keys, values, q_mask=None, kv_mask=None):
+        """ Multi-head scaled dot-product attention, a.k.a full attention.
+        Args:
+            queries: [N, L, H, D]
+            keys: [N, S, H, D]
+            values: [N, S, H, D]
+            q_mask: [N, L]
+            kv_mask: [N, S]
+        Returns:
+            queried_values: (N, L, H, D)
+        """
+
+        # Compute the unnormalized attention and apply the masks
+        QK = torch.einsum("nlhd,nshd->nlsh", queries, keys)
+        if kv_mask is not None:
+            QK.masked_fill_(~(q_mask[:, :, None, None] * kv_mask[:, None, :, None]), float('-inf'))
+
+        # Compute the attention and the weighted average
+        softmax_temp = 1. / queries.size(3)**.5  # sqrt(D)
+        A = torch.softmax(softmax_temp * QK, dim=2)
+        if self.use_dropout:
+            A = self.dropout(A)
+
+        queried_values = torch.einsum("nlsh,nshd->nlhd", A, values)
+
+        return queried_values.contiguous()

vismatch/third_party/TopicFM/third_party/loftr/src/loftr/loftr_module/transformer.py

@@ -0,0 +1,101 @@
+import copy
+import torch
+import torch.nn as nn
+from .linear_attention import LinearAttention, FullAttention
+
+
+class LoFTREncoderLayer(nn.Module):
+    def __init__(self,
+                 d_model,
+                 nhead,
+                 attention='linear'):
+        super(LoFTREncoderLayer, self).__init__()
+
+        self.dim = d_model // nhead
+        self.nhead = nhead
+
+        # multi-head attention
+        self.q_proj = nn.Linear(d_model, d_model, bias=False)
+        self.k_proj = nn.Linear(d_model, d_model, bias=False)
+        self.v_proj = nn.Linear(d_model, d_model, bias=False)
+        self.attention = LinearAttention() if attention == 'linear' else FullAttention()
+        self.merge = nn.Linear(d_model, d_model, bias=False)
+
+        # feed-forward network
+        self.mlp = nn.Sequential(
+            nn.Linear(d_model*2, d_model*2, bias=False),
+            nn.ReLU(True),
+            nn.Linear(d_model*2, d_model, bias=False),
+        )
+
+        # norm and dropout
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+
+    def forward(self, x, source, x_mask=None, source_mask=None):
+        """
+        Args:
+            x (torch.Tensor): [N, L, C]
+            source (torch.Tensor): [N, S, C]
+            x_mask (torch.Tensor): [N, L] (optional)
+            source_mask (torch.Tensor): [N, S] (optional)
+        """
+        bs = x.size(0)
+        query, key, value = x, source, source
+
+        # multi-head attention
+        query = self.q_proj(query).view(bs, -1, self.nhead, self.dim)  # [N, L, (H, D)]
+        key = self.k_proj(key).view(bs, -1, self.nhead, self.dim)  # [N, S, (H, D)]
+        value = self.v_proj(value).view(bs, -1, self.nhead, self.dim)
+        message = self.attention(query, key, value, q_mask=x_mask, kv_mask=source_mask)  # [N, L, (H, D)]
+        message = self.merge(message.view(bs, -1, self.nhead*self.dim))  # [N, L, C]
+        message = self.norm1(message)
+
+        # feed-forward network
+        message = self.mlp(torch.cat([x, message], dim=2))
+        message = self.norm2(message)
+
+        return x + message
+
+
+class LocalFeatureTransformer(nn.Module):
+    """A Local Feature Transformer (LoFTR) module."""
+
+    def __init__(self, config):
+        super(LocalFeatureTransformer, self).__init__()
+
+        self.config = config
+        self.d_model = config['d_model']
+        self.nhead = config['nhead']
+        self.layer_names = config['layer_names']
+        encoder_layer = LoFTREncoderLayer(config['d_model'], config['nhead'], config['attention'])
+        self.layers = nn.ModuleList([copy.deepcopy(encoder_layer) for _ in range(len(self.layer_names))])
+        self._reset_parameters()
+
+    def _reset_parameters(self):
+        for p in self.parameters():
+            if p.dim() > 1:
+                nn.init.xavier_uniform_(p)
+
+    def forward(self, feat0, feat1, mask0=None, mask1=None):
+        """
+        Args:
+            feat0 (torch.Tensor): [N, L, C]
+            feat1 (torch.Tensor): [N, S, C]
+            mask0 (torch.Tensor): [N, L] (optional)
+            mask1 (torch.Tensor): [N, S] (optional)
+        """
+
+        assert self.d_model == feat0.size(2), "the feature number of src and transformer must be equal"
+
+        for layer, name in zip(self.layers, self.layer_names):
+            if name == 'self':
+                feat0 = layer(feat0, feat0, mask0, mask0)
+                feat1 = layer(feat1, feat1, mask1, mask1)
+            elif name == 'cross':
+                feat0 = layer(feat0, feat1, mask0, mask1)
+                feat1 = layer(feat1, feat0, mask1, mask0)
+            else:
+                raise KeyError
+
+        return feat0, feat1