diff --git a/tracking_re_id/tracking_re_id/KeyRe_ID_model.py b/tracking_re_id/tracking_re_id/KeyRe_ID_model.py
new file mode 100644
index 0000000..1aeb759
--- /dev/null
+++ b/tracking_re_id/tracking_re_id/KeyRe_ID_model.py
@@ -0,0 +1,298 @@
+import torch
+import torch.nn as nn
+import copy
+from .vit_ID import TransReID, Block
+from functools import partial
+from torch.nn import functional as F
+from .vit_ID import resize_pos_embed
+
+
+def TCSS(features, shift, b,t):
+    # aggregate features at patch level
+    features = features.view(b, features.size(1), t*features.size(2))
+    token = features[:, 0:1]
+
+    batchsize = features.size(0)
+    dim = features.size(-1)
+    
+    # shift the patches with amount=shift
+    features= torch.cat([features[:, shift:], features[:, 1:shift]], dim=1)
+    
+    # Patch Shuffling by 2 part
+    try:
+        features = features.view(batchsize, 2, -1, dim)
+    except:
+        features = torch.cat([features, features[:, -2:-1, :]], dim=1)
+        features = features.view(batchsize, 2, -1, dim)
+    
+    features = torch.transpose(features, 1, 2).contiguous()                                                                                                                                   
+    features = features.view(batchsize, -1, dim)
+    
+    return features, token    
+
+def weights_init_kaiming(m):
+    classname = m.__class__.__name__
+    if classname.find('Linear') != -1:
+        nn.init.kaiming_normal_(m.weight, a=0, mode='fan_out')
+        nn.init.constant_(m.bias, 0.0)
+    elif classname.find('Conv') != -1:
+        nn.init.kaiming_normal_(m.weight, a=0, mode='fan_in')
+        if m.bias is not None:
+            nn.init.constant_(m.bias, 0.0)
+    elif classname.find('BatchNorm') != -1:
+        if m.affine:
+            nn.init.constant_(m.weight, 1.0)
+            nn.init.constant_(m.bias, 0.0)
+
+def weights_init_classifier(m):
+    classname = m.__class__.__name__
+    if classname.find('Linear') != -1:
+        nn.init.normal_(m.weight, std=0.001)
+        if m.bias:
+            nn.init.constant_(m.bias, 0.0)
+
+
+class KeyRe_ID(nn.Module):
+    def __init__(self, num_classes, camera_num, pretrainpath):
+        super(KeyRe_ID, self).__init__()
+        self.in_planes = 768
+        self.num_classes = num_classes
+        
+        self.base =TransReID(
+        img_size=[256, 128], patch_size=16, stride_size=[16, 16], embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,\
+        camera=camera_num,  drop_path_rate=0.1, drop_rate=0.0, attn_drop_rate=0.0,norm_layer=partial(nn.LayerNorm, eps=1e-6), cam_lambda=3.0)
+        
+        # state_dict = torch.load(pretrainpath, map_location='cpu')
+        # self.base.load_param(state_dict,load=True)
+        if pretrainpath:
+            state_dict = torch.load(pretrainpath, map_location='cpu', weights_only=False)
+            self.base.load_param(state_dict, load=True)
+        
+        #-------------------Global Branch-------------
+        block= self.base.blocks[-1]
+        layer_norm = self.base.norm
+        self.b1 = nn.Sequential(
+            copy.deepcopy(block),
+            copy.deepcopy(layer_norm)
+        )
+        
+        self.bottleneck = nn.BatchNorm1d(self.in_planes)
+        self.bottleneck.bias.requires_grad_(False)
+        self.bottleneck.apply(weights_init_kaiming)
+        self.classifier = nn.Linear(self.in_planes, self.num_classes, bias=False)
+        self.classifier.apply(weights_init_classifier)
+        
+        #-------------------Local Branch-------------
+        # building local video stream
+        dpr = [x.item() for x in torch.linspace(0, 0, 12)]  # stochastic depth decay rule
+        
+        self.block1 = Block(
+                dim=3072, num_heads=12, mlp_ratio=4, qkv_bias=True, qk_scale=None,
+                drop=0, attn_drop=0, drop_path=dpr[11], norm_layer=partial(nn.LayerNorm, eps=1e-6))
+        self.b2 = nn.Sequential(
+            self.block1,
+            nn.LayerNorm(3072)  # copy.deepcopy(layer_norm)
+        )
+        
+        self.bottleneck_1 = nn.BatchNorm1d(3072)
+        self.bottleneck_1.bias.requires_grad_(False)
+        self.bottleneck_1.apply(weights_init_kaiming)
+        self.bottleneck_2 = nn.BatchNorm1d(3072)
+        self.bottleneck_2.bias.requires_grad_(False)
+        self.bottleneck_2.apply(weights_init_kaiming)
+        self.bottleneck_3 = nn.BatchNorm1d(3072)
+        self.bottleneck_3.bias.requires_grad_(False)
+        self.bottleneck_3.apply(weights_init_kaiming)
+        self.bottleneck_4 = nn.BatchNorm1d(3072)
+        self.bottleneck_4.bias.requires_grad_(False)
+        self.bottleneck_4.apply(weights_init_kaiming)
+        self.bottleneck_5 = nn.BatchNorm1d(3072)
+        self.bottleneck_5.bias.requires_grad_(False)
+        self.bottleneck_5.apply(weights_init_kaiming)
+        self.bottleneck_6 = nn.BatchNorm1d(3072)
+        self.bottleneck_6.bias.requires_grad_(False)
+        self.bottleneck_6.apply(weights_init_kaiming)
+
+        self.classifier_1 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_1.apply(weights_init_classifier)
+        self.classifier_2 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_2.apply(weights_init_classifier)
+        self.classifier_3 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_3.apply(weights_init_classifier)
+        self.classifier_4 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_4.apply(weights_init_classifier)
+        self.classifier_5 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_5.apply(weights_init_classifier)
+        self.classifier_6 = nn.Linear(3072, self.num_classes, bias=False)
+        self.classifier_6.apply(weights_init_classifier)
+        
+        #-------------------video attention-------------
+        self.middle_dim = 256  # middle layer dimension
+        self.attention_conv = nn.Conv2d(self.in_planes, self.middle_dim, [1,1])  # 7,4 cooresponds to 224, 112 input image size
+        self.attention_tconv = nn.Conv1d(self.middle_dim, 1, 3, padding=1)
+        self.attention_conv.apply(weights_init_kaiming) 
+        self.attention_tconv.apply(weights_init_kaiming) 
+        #------------------------------------------
+        self.shift_num = 5
+        self.part = 6
+        self.rearrange=True 
+        
+    def forward(self, x, heatmaps, label=None, cam_label= None, view_label=None):  # label is unused if self.cos_layer == 'no'
+        b = x.size(0)
+        t = x.size(1)
+        
+        x = x.view(x.size(0)*x.size(1), x.size(2), x.size(3), x.size(4))
+        features = self.base(x, cam_label=cam_label)
+        
+        #-------------------Global Branch-------------
+        b1_feat = self.b1(features)  # [64, 129, 3072]
+        global_feat = b1_feat[:, 0]
+        
+        global_feat = global_feat.unsqueeze(dim=2).unsqueeze(dim=3)
+        a = F.relu(self.attention_conv(global_feat))
+        a = a.view(b, t, self.middle_dim)
+        a = a.permute(0,2,1)
+        a = F.relu(self.attention_tconv(a))
+        a = a.view(b, t)
+        a_vals = a 
+        
+        a = F.softmax(a, dim=1)
+        x = global_feat.view(b, t, -1)
+        a = torch.unsqueeze(a, -1)
+        a = a.expand(b, t, self.in_planes)
+        att_x = torch.mul(x,a)
+        att_x = torch.sum(att_x, 1)
+        
+        global_feat = att_x.view(b, self.in_planes)
+        feat = self.bottleneck(global_feat)
+        
+        #-------------------Local Branch-------------
+        # Heatmap Processing
+        heatmaps = heatmaps.view(b*t, 6, 256, 128)  # [B*T, 6, 256, 128]
+        heatmap_patches = F.unfold(heatmaps, kernel_size=16, stride=16)  # [B*T, 6*16*16, 128]
+        heatmap_patches = heatmap_patches.view(b*t, 6, 16*16, 128).mean(dim=2)  # [B*T, 6, 128]
+        heatmap_weights = heatmap_patches.transpose(1, 2)  # [B*T, 128, 6]
+        heatmap_weights = heatmap_weights.view(b, t, 128, 6).mean(dim=1)  # [B, 128, 6]
+        
+        # Temporal clip shift and shuffled
+        x ,token = TCSS(features, self.shift_num, b, t)  
+        patch_feats = x
+        
+        # Part 1: Head
+        part1_weight = heatmap_weights[:, :, 0].unsqueeze(-1)
+        part1 = patch_feats * part1_weight
+        part1 = self.b2(torch.cat((token, part1), dim=1))
+        part1_f = part1[:, 0]
+        
+        # Part 2: Torso
+        part2_weight = heatmap_weights[:, :, 1].unsqueeze(-1)
+        part2 = patch_feats * part2_weight
+        part2 = self.b2(torch.cat((token, part2), dim=1))
+        part2_f = part2[:, 0]
+        
+        # Part 3: Left Arm
+        part3_weight = heatmap_weights[:, :, 2].unsqueeze(-1)
+        part3 = patch_feats * part3_weight
+        part3 = self.b2(torch.cat((token, part3), dim=1))
+        part3_f = part3[:, 0]
+        
+        # Part 4: Right Arm
+        part4_weight = heatmap_weights[:, :, 3].unsqueeze(-1)
+        part4 = patch_feats * part4_weight
+        part4 = self.b2(torch.cat((token, part4), dim=1))
+        part4_f = part4[:, 0]
+        
+        # Part 5: Left Leg
+        part5_weight = heatmap_weights[:, :, 4].unsqueeze(-1)
+        part5 = patch_feats * part5_weight
+        part5 = self.b2(torch.cat((token, part5), dim=1))
+        part5_f = part5[:, 0]
+        
+        # Part 6: Right Leg
+        part6_weight = heatmap_weights[:, :, 5].unsqueeze(-1)
+        part6 = patch_feats * part6_weight
+        part6 = self.b2(torch.cat((token, part6), dim=1))
+        part6_f = part6[:, 0]
+        
+        # Apply batch normalization
+        part1_bn = self.bottleneck_1(part1_f)
+        part2_bn = self.bottleneck_2(part2_f)
+        part3_bn = self.bottleneck_3(part3_f)
+        part4_bn = self.bottleneck_4(part4_f)
+        part5_bn = self.bottleneck_5(part5_f)
+        part6_bn = self.bottleneck_6(part6_f)
+        
+        if self.training:
+            Global_ID = self.classifier(feat)
+            Local_ID1 = self.classifier_1(part1_bn)
+            Local_ID2 = self.classifier_2(part2_bn)
+            Local_ID3 = self.classifier_3(part3_bn)
+            Local_ID4 = self.classifier_4(part4_bn)
+            Local_ID5 = self.classifier_5(part5_bn)
+            Local_ID6 = self.classifier_6(part6_bn)
+                
+            return [Global_ID, Local_ID1, Local_ID2, Local_ID3, Local_ID4, Local_ID5, Local_ID6],\
+                [global_feat, part1_f, part2_f, part3_f, part4_f, part5_f, part6_f], a_vals 
+        else:
+            return torch.cat([feat, part1_bn/self.part, part2_bn/self.part, part3_bn/self.part, 
+                              part4_bn/self.part, part5_bn/self.part, part6_bn/self.part], dim=1)
+
+    def load_param(self, trained_path, load=False):
+        print("Run load_param")
+        if not load:
+            param_dict = torch.load(trained_path, map_location='cpu', weights_only=False)
+        else:
+            param_dict = trained_path
+
+        if 'model' in param_dict:
+            param_dict = param_dict['model']
+        if 'state_dict' in param_dict:
+            param_dict = param_dict['state_dict']
+
+        model_dict = self.state_dict()  # Get the state_dict of the current model
+        new_param_dict = {}
+
+        for k, v in param_dict.items():
+            if 'head' in k or 'dist' in k:
+                continue 
+
+            # Patch embedding Conv-based transformation processing
+            if 'patch_embed.proj.weight' in k and len(v.shape) < 4:
+                O, I, H, W = self.base.patch_embed.proj.weight.shape
+                v = v.reshape(O, -1, H, W)
+            # Resize Positional Embedding
+            elif k == 'pos_embed' and v.shape != self.base.pos_embed.shape:
+                v = resize_pos_embed(v, self.base.pos_embed, self.base.patch_embed.num_y, self.base.patch_embed.num_x)
+
+            # Handling `base.` prefix
+            new_k = k
+            if k.startswith("base.") and k[5:] in model_dict:
+                new_k = k[5:]  # Remove base.
+            elif not k.startswith("base.") and ("base." + k) in model_dict:
+                new_k = "base." + k  # Add base.
+
+            if new_k in ['Cam', 'base.Cam'] and new_k in model_dict:
+                expected_shape = model_dict[new_k].shape  # Cam size that the current model expects
+                print(f"[Before Resizing] {new_k}: {v.shape} -> Expected: {expected_shape}") 
+                
+                if v.shape[0] > expected_shape[0]:  # Keep only the front part if the size is larger
+                    v = v[:expected_shape[0], :, :]
+                elif v.shape[0] < expected_shape[0]:  # Create a new tensor for smaller sizes
+                    new_v = torch.randn(expected_shape)  # Random initialization (other values are possible)
+                    new_v[:v.shape[0], :, :] = v  # Keep existing values
+                    v = new_v
+
+                print(f"[After Resizing] {new_k}: {v.shape}")  # Confirm after changing the size
+                new_param_dict[new_k] = v
+                continue
+
+            # Update only if Shape fits
+            if new_k in model_dict and model_dict[new_k].shape == v.shape:
+                new_param_dict[new_k] = v
+
+        # Finally, update the state_dict
+        model_dict.update(new_param_dict)
+        self.load_state_dict(model_dict, strict=False)
+        print("Checkpoint loaded successfully.")
+
+    
\ No newline at end of file
diff --git a/tracking_re_id/tracking_re_id/reid_node.py b/tracking_re_id/tracking_re_id/reid_node.py
index cc02cfc..b9b31ec 100644
--- a/tracking_re_id/tracking_re_id/reid_node.py
+++ b/tracking_re_id/tracking_re_id/reid_node.py
@@ -31,7 +31,6 @@ Pipeline (per frame)
 Parameters
 ──────────
   weights_path    str    path to iLIDSVIDbest_CMC.pth        (required)
-  keyreID_path    str    path to KeyRe-ID source directory
   num_classes     int    training split size (150 for iLIDS-VID split-0)
   camera_num      int    cameras in training set (2 for iLIDS-VID)
   device          str    'cuda:0' or 'cpu'
@@ -44,7 +43,6 @@ Parameters
 """
 
 import os
-import sys
 import time
 import colorsys
 
@@ -140,8 +138,6 @@ class ReIDNode(Node):
                                os.path.join(
                                    get_package_share_directory('tracking_re_id'),
                                    'weights', 'iLIDSVIDbest_CMC.pth'))
-        self.declare_parameter('keyreID_path',
-                               os.path.expanduser('~/KeyRe-ID'))
         self.declare_parameter('num_classes',  150)
         self.declare_parameter('camera_num',   2)
         self.declare_parameter('device',       'cuda:0')
@@ -153,7 +149,6 @@ class ReIDNode(Node):
         self.declare_parameter('headless',     False)
 
         weights_path      = self.get_parameter('weights_path').value
-        keyreID_path      = self.get_parameter('keyreID_path').value
         num_classes       = self.get_parameter('num_classes').value
         camera_num        = self.get_parameter('camera_num').value
         device_str        = self.get_parameter('device').value
@@ -175,14 +170,7 @@ class ReIDNode(Node):
         self.get_logger().info('MMPose loaded.')
 
         # ── KeyRe-ID ─────────────────────────────────────────────────────────
-        if keyreID_path not in sys.path:
-            sys.path.insert(0, keyreID_path)
-        try:
-            from KeyRe_ID_model import KeyRe_ID  # noqa: PLC0415
-        except ImportError as exc:
-            self.get_logger().fatal(
-                f'Cannot import KeyRe_ID_model from {keyreID_path}: {exc}')
-            raise
+        from .KeyRe_ID_model import KeyRe_ID  # noqa: PLC0415
 
         self.get_logger().info(f'Loading KeyRe-ID weights from {weights_path} …')
         self._model = KeyRe_ID(
diff --git a/tracking_re_id/tracking_re_id/vit_ID.py b/tracking_re_id/tracking_re_id/vit_ID.py
new file mode 100644
index 0000000..1e89598
--- /dev/null
+++ b/tracking_re_id/tracking_re_id/vit_ID.py
@@ -0,0 +1,352 @@
+import math
+from itertools import repeat
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import collections.abc
+
+
+# From PyTorch internals
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+    return parse
+
+IMAGENET_DEFAULT_MEAN = (0.485, 0.456, 0.406)
+IMAGENET_DEFAULT_STD = (0.229, 0.224, 0.225)
+to_2tuple = _ntuple(2)
+
+def drop_path(x, drop_prob: float = 0., training: bool = False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+
+    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
+    changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
+    'survival rate' as the argument.
+
+    """
+    if drop_prob == 0. or not training:
+        return x
+    keep_prob = 1 - drop_prob
+    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
+    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
+    random_tensor.floor_()  # binarize
+    output = x.div(keep_prob) * random_tensor
+    return output
+
+class DropPath(nn.Module):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+class Mlp(nn.Module):
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
+        self.scale = qk_scale or head_dim ** -0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]   # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+    
+class Block(nn.Module):
+    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+    def forward(self, x):
+        x = x + self.drop_path(self.attn(self.norm1(x)))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+class PatchEmbed(nn.Module):
+    """ Image to Patch Embedding"""
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.num_patches = num_patches
+
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        # FIXME look at relaxing size constraints
+        assert H == self.img_size[0] and W == self.img_size[1], \
+            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+        x = self.proj(x).flatten(2).transpose(1, 2)
+        return x
+
+class PatchEmbed_overlap(nn.Module):
+    """ Image to Patch Embedding with overlapping patches"""
+    def __init__(self, img_size=224, patch_size=16, stride_size=20, in_chans=3, embed_dim=768):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        stride_size_tuple = to_2tuple(stride_size)
+        self.num_x = (img_size[1] - patch_size[1]) // stride_size_tuple[1] + 1
+        self.num_y = (img_size[0] - patch_size[0]) // stride_size_tuple[0] + 1
+        print('using stride: {}, and patch number is num_y{} * num_x{}'.format(stride_size, self.num_y, self.num_x))
+        num_patches = self.num_x * self.num_y
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.num_patches = num_patches
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=stride_size)
+        
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
+                m.weight.data.normal_(0, math.sqrt(2. / n))
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.InstanceNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        # FIXME look at relaxing size constraints
+        assert H == self.img_size[0] and W == self.img_size[1], \
+            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+        x = self.proj(x)
+        x = x.flatten(2).transpose(1, 2) # [64, 8, 768]
+        
+        return x
+
+class TransReID(nn.Module):
+    """ Transformer-based Object Re-Identification"""
+    def __init__(self, img_size=224, patch_size=16, stride_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
+                 num_heads=12, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0., camera=0, 
+                 drop_path_rate=0., hybrid_backbone=None, norm_layer=nn.LayerNorm, cam_lambda =3.0):
+        super().__init__()
+        self.num_classes = num_classes
+        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
+        self.cam_num = camera
+        self.cam_lambda = cam_lambda
+
+        self.patch_embed = PatchEmbed_overlap(img_size=img_size, patch_size=patch_size, stride_size=stride_size, in_chans=in_chans,embed_dim=embed_dim)
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
+        self.Cam = nn.Parameter(torch.zeros(camera, 1, embed_dim))
+
+        trunc_normal_(self.Cam, std=.02)
+        self.pos_drop = nn.Dropout(p=drop_rate)
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
+
+        self.blocks = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer)
+            for i in range(depth)])
+
+        self.norm = norm_layer(embed_dim)
+
+        # Classifier head
+        self.fc = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()
+        trunc_normal_(self.cls_token, std=.02)
+        trunc_normal_(self.pos_embed, std=.02)
+
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'pos_embed', 'cls_token'}
+
+    def get_classifier(self):
+        return self.head
+
+    def reset_classifier(self, num_classes, global_pool=''):
+        self.num_classes = num_classes
+        self.fc = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()
+
+    def forward_features(self, x, camera_id):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+        
+        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+        x = x + self.pos_embed + self.cam_lambda * self.Cam[camera_id]
+        x = self.pos_drop(x)
+
+        for blk in self.blocks[:-1]:
+                x = blk(x)
+        return x
+        
+    def forward(self, x, cam_label=None):
+        x = self.forward_features(x, cam_label)
+        return x
+
+    def load_param(self, model_path, load=False):
+        print("Run load_param")
+        if not load:
+            param_dict = torch.load(model_path, map_location='cpu', weights_only=False)
+        else:
+            param_dict = model_path
+
+        if 'model' in param_dict:
+            param_dict = param_dict['model']
+        if 'state_dict' in param_dict:
+            param_dict = param_dict['state_dict']
+
+        model_dict = self.state_dict()
+        new_param_dict = {}
+
+        for k, v in param_dict.items():
+            if 'head' in k or 'dist' in k:
+                continue
+
+            if k in ['Cam', 'base.Cam'] and k in model_dict:
+                expected_shape = model_dict[k].shape
+                if v.shape[0] > expected_shape[0]:
+                    print(f"⚠️ Resizing '{k}' from {v.shape} to {expected_shape}")
+                    v = v[:expected_shape[0], :, :]
+                elif v.shape[0] < expected_shape[0]:
+                    print(f"⚠️ Expanding '{k}' from {v.shape} to {expected_shape}")
+                    new_v = torch.randn(expected_shape)
+                    new_v[:v.shape[0], :, :] = v
+                    v = new_v
+                new_param_dict[k] = v
+                continue
+
+            if k in model_dict and model_dict[k].shape == v.shape:
+                new_param_dict[k] = v
+
+        model_dict.update(new_param_dict)
+        self.load_state_dict(model_dict, strict=False)
+        print("✅ Checkpoint loaded successfully.")
+
+def resize_pos_embed(posemb, posemb_new, hight, width):
+    # Rescale the grid of position embeddings when loading from state_dict. Adapted from
+    # https://github.com/google-research/vision_transformer/blob/00883dd691c63a6830751563748663526e811cee/vit_jax/checkpoint.py#L224
+    ntok_new = posemb_new.shape[1]
+
+    posemb_token, posemb_grid = posemb[:, :1], posemb[0, 1:]
+    ntok_new -= 1
+
+    gs_old = int(math.sqrt(len(posemb_grid)))
+    print('Resized position embedding from size:{} to size: {} with height:{} width: {}'.format(posemb.shape, posemb_new.shape, hight, width))
+    posemb_grid = posemb_grid.reshape(1, gs_old, gs_old, -1).permute(0, 3, 1, 2)
+    posemb_grid = F.interpolate(posemb_grid, size=(hight, width), mode='bilinear')
+    posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, hight * width, -1)
+    posemb = torch.cat([posemb_token, posemb_grid], dim=1)
+    return posemb
+
+def _no_grad_trunc_normal_(tensor, mean, std, a, b):
+    # Cut & paste from PyTorch official master until it's in a few official releases - RW
+    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
+    def norm_cdf(x):
+        # Computes standard normal cumulative distribution function
+        return (1. + math.erf(x / math.sqrt(2.))) / 2.
+
+    if (mean < a - 2 * std) or (mean > b + 2 * std):
+        print("mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
+                      "The distribution of values may be incorrect.",)
+
+    with torch.no_grad():
+        # Values are generated by using a truncated uniform distribution and
+        # then using the inverse CDF for the normal distribution.
+        # Get upper and lower cdf values
+        l = norm_cdf((a - mean) / std)
+        u = norm_cdf((b - mean) / std)
+
+        # Uniformly fill tensor with values from [l, u], then translate to
+        # [2l-1, 2u-1].
+        tensor.uniform_(2 * l - 1, 2 * u - 1)
+
+        # Use inverse cdf transform for normal distribution to get truncated
+        # standard normal
+        tensor.erfinv_()
+
+        # Transform to proper mean, std
+        tensor.mul_(std * math.sqrt(2.))
+        tensor.add_(mean)
+
+        # Clamp to ensure it's in the proper range
+        tensor.clamp_(min=a, max=b)
+        return tensor
+
+
+def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
+    # type: (Tensor, float, float, float, float) -> Tensor
+    r"""Fills the input Tensor with values drawn from a truncated
+    normal distribution. The values are effectively drawn from the
+    normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)`
+    with values outside :math:`[a, b]` redrawn until they are within
+    the bounds. The method used for generating the random values works
+    best when :math:`a \leq \text{mean} \leq b`.
+    Args:
+        tensor: an n-dimensional `torch.Tensor`
+        mean: the mean of the normal distribution
+        std: the standard deviation of the normal distribution
+        a: the minimum cutoff value
+        b: the maximum cutoff value
+    Examples:
+        >>> w = torch.empty(3, 5)
+        >>> nn.init.trunc_normal_(w)
+    """
+    return _no_grad_trunc_normal_(tensor, mean, std, a, b)