Upload V_JEPA

config.json

@@ -7,7 +7,6 @@
     "AutoModel": "model.V_JEPA"
   },
   "ckpt_path": "/home/linanxi/V-JEPA/ckpt/vitl16.pth.tar",
-  "device": "cuda:2",
   "model_type": "v-jepa",
   "torch_dtype": "float32",
   "transformers_version": "4.47.0",

model.py

@@ -1,10 +1,10 @@
+from unittest.mock import Base
 from transformers import PretrainedConfig, PreTrainedModel, AutoModel, AutoConfig
 from collections import OrderedDict
-from vision_transformer import vit_huge_16, vit_large_16
+from transformers.modeling_outputs import BaseModelOutput
+from .vision_transformer import vit_huge_16, vit_large_16
 import torch
 
-device = 'cuda:2'
-
 class JEPAConfig(PretrainedConfig):
     model_type = "v-jepa"
 
@@ -12,13 +12,12 @@ class JEPAConfig(PretrainedConfig):
         self, 
         vit_type: str = 'vit_large_16', 
         ckpt_path: str = None, 
-        device: str = 'cuda' if torch.cuda.is_available() else 'cpu',
         **kwargs
         ):
         super().__init__(**kwargs)
         self.vit_type = vit_type
         self.ckpt_path = ckpt_path
-        self.device = device
+        # print(f'V-JEPA Config: {self.vit_type}, {self.ckpt_path}, {self.device}')
 
 
 class V_JEPA(PreTrainedModel):
@@ -28,9 +27,9 @@ class V_JEPA(PreTrainedModel):
         super().__init__(config)
         self.config = config
         if config.vit_type == 'vit_large_16':
-            self.model = vit_large_16().to(config.device)
+            self.model = vit_large_16()
         elif config.vit_type == 'vit_huge_16':
-            self.model = vit_huge_16().to(config.device)
+            self.model = vit_huge_16()
         else:
             raise ValueError(f"Unsupported vit_type: {config.vit_type}")
 
@@ -39,14 +38,42 @@ class V_JEPA(PreTrainedModel):
 
     def load_checkpoint(self, ckpt_path):
         state_dict = OrderedDict()
-        ckpt = torch.load(ckpt_path, weights_only=False, map_location=self.config.device)['encoder']
+        ckpt = torch.load(ckpt_path, weights_only=False, map_location='cpu')['encoder']
         for k, v in ckpt.items():
             new_key = k.split('.', 1)[-1]
             state_dict[new_key] = v
         self.model.load_state_dict(state_dict, strict=False)
         print("Checkpoint loaded successfully")
 
-    def forward(self, x):
-        return self.model(x)
+    def forward(self, x: torch.tensor):
+        """forward pass 
+        Args:
+            x (torch.tensor): Shape (B, N, C, H, W) or (N, C, H, W)
+        Returns:
+            torch.tensor: Shape (B, N, hidden_size)
+        """
+        # if len(x.shape) == 5 and x.shape[1] == 9:
+        #     x_8T = x[:, :8, :, :, :]
+        #     x_1T = x[:, 8, :, :, :].unsqueeze(1)
+        #     y_8T = self.forward(x_8T).last_hidden_state
+        #     y_1T = self.forward(x_1T).last_hidden_state
+        #     output = torch.cat((y_8T, y_1T), dim=1)
+        #     return BaseModelOutput(last_hidden_state=output)
+        
+        if len(x.shape) == 4:
+            x = x.unsqueeze(0)
+        B, N, C, H, W = x.shape
+        x = x.permute(0, 2, 1, 3, 4)    # Shape(B, C, N, H, W)
+        output = self.model(x)      
+        output = output.view(B, N, 98, -1) # Shape(B*N, 98, hidden_size)
+        output = output.mean(dim=2)         # Shape(B*N, hidden_size)
+        # print("output shape: ", output.shape)
+        return BaseModelOutput(last_hidden_state=output)
 
 
+if __name__ == "__main__":
+    config = JEPAConfig()
+    model = V_JEPA(config)
+    x = torch.randn(2, 8, 3, 224, 224)
+    output = model(x)
+    print(output.shape) # (16, 1024)

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c81ae01decbe7a322042b7393e75397ff795e2edd9547a8847c025e4c259f9f5
+oid sha256:88ebddcc1c98353e2c8a6c4a55800a9c36e6bd609134c446acf06a2b433e074b
 size 818904440

modules.py

@@ -0,0 +1,183 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+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.,
+        use_sdpa=True
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        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_prob = proj_drop
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.use_sdpa = use_sdpa
+
+    def forward(self, x, mask=None):
+        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]  # [B, num_heads, N, D]
+
+        if self.use_sdpa:
+            with torch.backends.cuda.sdp_kernel():
+                x = F.scaled_dot_product_attention(q, k, v, dropout_p=self.proj_drop_prob)
+                attn = None
+        else:
+            attn = (q @ k.transpose(-2, -1)) * self.scale  # [B, num_heads, D, D]
+            attn = attn.softmax(dim=-1)
+            attn = self.attn_drop(attn)
+            x = (attn @ v)
+        x = x.transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x, attn
+
+
+class Block(nn.Module):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.,
+        attn_drop=0.,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+        grid_size=None,
+        grid_depth=None,
+    ):
+        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)
+
+        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, return_attention=False, mask=None):
+        y, attn = self.attn(self.norm1(x), mask=mask)
+        if return_attention:
+            return attn
+        x = x + y
+        x = x + self.mlp(self.norm2(x))
+        return x
+
+
+class CrossAttention(nn.Module):
+    def __init__(
+        self,
+        dim,
+        num_heads=12,
+        qkv_bias=False,
+        use_sdpa=True
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim ** -0.5
+        self.q = nn.Linear(dim, dim, bias=qkv_bias)
+        self.kv = nn.Linear(dim, int(dim*2), bias=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+        self.use_sdpa = use_sdpa
+
+    def forward(self, q, x):
+        B, n, C = q.shape
+        q = self.q(q).reshape(B, n, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
+
+        B, N, C = x.shape
+        kv = self.kv(x).reshape(B, N, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        k, v = kv[0], kv[1]  # (batch_size, num_heads, seq_len, feature_dim_per_head)
+
+        if self.use_sdpa:
+            with torch.backends.cuda.sdp_kernel():
+                q = F.scaled_dot_product_attention(q, k, v)
+        else:
+            xattn = (q @ k.transpose(-2, -1)) * self.scale
+            xattn = xattn.softmax(dim=-1)  # (batch_size, num_heads, query_len, seq_len)
+            q = (xattn @ v)
+
+        q = q.transpose(1, 2).reshape(B, n, C)
+        q = self.proj(q)
+    
+        return q
+
+
+class CrossAttentionBlock(nn.Module):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.,
+        qkv_bias=False,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm
+    ):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.xattn = CrossAttention(dim, num_heads=num_heads, qkv_bias=qkv_bias)
+        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)
+
+    def forward(self, q, x):
+        y = self.xattn(q, self.norm1(x))
+        q = q + y
+        q = q + self.mlp(self.norm2(q))
+        return q

patch_embed.py

@@ -0,0 +1,57 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import torch.nn as nn
+
+
+class PatchEmbed(nn.Module):
+    """
+    Image to Patch Embedding
+    """
+    def __init__(
+        self,
+        patch_size=16,
+        in_chans=3,
+        embed_dim=768
+    ):
+        super().__init__()
+        self.patch_size = patch_size
+        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
+        x = self.proj(x).flatten(2).transpose(1, 2)
+        return x
+
+
+class PatchEmbed3D(nn.Module):
+    """
+    Image to Patch Embedding
+    """
+
+    def __init__(
+        self,
+        patch_size=16,
+        tubelet_size=2,
+        in_chans=3,
+        embed_dim=768,
+    ):
+        super().__init__()
+        self.patch_size = patch_size
+        self.tubelet_size = tubelet_size
+
+        self.proj = nn.Conv3d(
+            in_channels=in_chans,
+            out_channels=embed_dim,
+            kernel_size=(tubelet_size, patch_size, patch_size),
+            stride=(tubelet_size, patch_size, patch_size),
+        )
+
+    def forward(self, x, **kwargs):
+        B, C, T, H, W = x.shape
+        x = self.proj(x).flatten(2).transpose(1, 2)
+        return x

pos_embs.py

@@ -0,0 +1,99 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import numpy as np
+
+
+def get_3d_sincos_pos_embed(
+    embed_dim,
+    grid_size,
+    grid_depth,
+    cls_token=False,
+    uniform_power=False
+):
+    """
+    grid_size: int of the grid height and width
+    grid_depth: int of the grid depth
+    returns:
+        pos_embed: [grid_depth*grid_size*grid_size, embed_dim] (w/o cls_token)
+                or [1+grid_depth*grid_size*grid_size, embed_dim] (w/ cls_token)
+    """
+    grid_d = np.arange(grid_depth, dtype=float)
+    grid_h = np.arange(grid_size, dtype=float)
+    grid_w = np.arange(grid_size, dtype=float)
+    grid_h, grid_d, grid_w = np.meshgrid(grid_h, grid_d, grid_w)  # order of meshgrid is very important for indexing as [d,h,w]
+
+    if not uniform_power:
+        h_embed_dim = embed_dim // 4
+        w_embed_dim = embed_dim // 4
+        d_embed_dim = embed_dim // 2
+    else:
+        h_embed_dim = w_embed_dim = d_embed_dim = int(np.ceil(embed_dim/6)*2)
+
+    emb_h = get_1d_sincos_pos_embed_from_grid(h_embed_dim, grid_h)  # (T*H*W, D1)
+    emb_w = get_1d_sincos_pos_embed_from_grid(w_embed_dim, grid_w)  # (T*H*W, D2)
+    emb_d = get_1d_sincos_pos_embed_from_grid(d_embed_dim, grid_d)  # (T*H*W, D3)
+    pos_embed = np.concatenate([emb_d, emb_h, emb_w], axis=1)
+    pos_embed = pos_embed[:, :embed_dim]
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+
+def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
+    """
+    grid_size: int of the grid height and width
+    returns:
+        pos_embed: [grid_size*grid_size, embed_dim] (w/o cls_token)
+                or [1+grid_size*grid_size, embed_dim] (w/ cls_token)
+    """
+    grid_h = np.arange(grid_size, dtype=float)
+    grid_w = np.arange(grid_size, dtype=float)
+    grid_w, grid_h = np.meshgrid(grid_w, grid_h)  # order of meshgrid is very important for indexing as [h, w]
+
+    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid_h)  # (H*W, D/2)
+    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid_w)  # (H*W, D/2)
+    pos_embed = np.concatenate([emb_h, emb_w], axis=1)  # (H*W, D)
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+
+def get_1d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
+    """
+    embed_dim: output dimension for each position
+    grid_size: int of the grid length
+    returns:
+        pos_embed: [grid_size, embed_dim] (w/o cls_token)
+                or [1+grid_size, embed_dim] (w/ cls_token)
+    """
+    grid = np.arange(grid_size, dtype=float)
+    pos_embed = get_1d_sincos_pos_embed_from_grid(embed_dim, grid)
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+
+def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
+    """
+    embed_dim: output dimension for each position
+    pos: a list of positions to be encoded: size (M,)
+    returns: (M, D)
+    """
+    assert embed_dim % 2 == 0
+    omega = np.arange(embed_dim // 2, dtype=float)
+    omega /= embed_dim / 2.
+    omega = 1. / 10000**omega   # (D/2,)
+
+    pos = pos.reshape(-1)   # (M,)
+    out = np.einsum('m,d->md', pos, omega)   # (M, D/2), outer product
+
+    emb_sin = np.sin(out)  # (M, D/2)
+    emb_cos = np.cos(out)  # (M, D/2)
+
+    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
+    return emb

tensors.py

@@ -0,0 +1,71 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import math
+
+import torch
+
+from logging import getLogger
+
+logger = getLogger()
+
+
+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.
+
+    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
+    return _no_grad_trunc_normal_(tensor, mean, std, a, b)
+
+
+def apply_masks(x, masks):
+    """
+    :param x: tensor of shape [B (batch-size), N (num-patches), D (feature-dim)]
+    :param masks: list of tensors containing indices of patches [0,N) to keep
+    """
+    all_x = []
+    for m in masks:
+        mask_keep = m.unsqueeze(-1).repeat(1, 1, x.size(-1))
+        all_x += [torch.gather(x, dim=1, index=mask_keep)]
+    return torch.cat(all_x, dim=0)
+
+
+def repeat_interleave_batch(x, B, repeat):
+    N = len(x) // B
+    x = torch.cat([
+        torch.cat([x[i*B:(i+1)*B] for _ in range(repeat)], dim=0)
+        for i in range(N)
+    ], dim=0)
+    return x

utils.py

@@ -0,0 +1,23 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import torch
+
+
+def apply_masks(x, masks, concat=True):
+    """
+    :param x: tensor of shape [B (batch-size), N (num-patches), D (feature-dim)]
+    :param masks: list of tensors of shape [B, K] containing indices of K patches in [N] to keep
+    """
+    all_x = []
+    for m in masks:
+        mask_keep = m.unsqueeze(-1).repeat(1, 1, x.size(-1))
+        all_x += [torch.gather(x, dim=1, index=mask_keep)]
+    if not concat:
+        return all_x
+
+    return torch.cat(all_x, dim=0)

vision_transformer.py

@@ -0,0 +1,324 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+#
+
+import math
+from functools import partial
+
+import torch
+import torch.nn as nn
+
+from .patch_embed import PatchEmbed, PatchEmbed3D
+from .modules import Block
+from .pos_embs import get_2d_sincos_pos_embed, get_3d_sincos_pos_embed
+from .tensors import trunc_normal_
+from .utils import apply_masks
+
+
+class VisionTransformer(nn.Module):
+    """ Vision Transformer """
+    def __init__(
+        self,
+        img_size=224,
+        patch_size=16,
+        num_frames=1,
+        tubelet_size=2,
+        in_chans=3,
+        embed_dim=768,
+        depth=12,
+        num_heads=12,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        norm_layer=nn.LayerNorm,
+        init_std=0.02,
+        out_layers=None,
+        uniform_power=False,
+        **kwargs
+    ):
+        super().__init__()
+        self.num_features = self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        self.out_layers = out_layers
+
+        self.input_size = img_size
+        self.patch_size = patch_size
+
+        self.num_frames = num_frames
+        self.tubelet_size = tubelet_size
+        self.is_video = num_frames > 1
+
+        grid_size = self.input_size // self.patch_size
+        grid_depth = self.num_frames // self.tubelet_size
+
+        # Tokenize pixels with convolution
+        if self.is_video:
+            self.patch_embed = PatchEmbed3D(
+                patch_size=patch_size,
+                tubelet_size=tubelet_size,
+                in_chans=in_chans,
+                embed_dim=embed_dim)
+            self.num_patches = (
+                (num_frames // tubelet_size)
+                * (img_size // patch_size)
+                * (img_size // patch_size)
+            )
+        else:
+            self.patch_embed = PatchEmbed(
+                patch_size=patch_size,
+                in_chans=in_chans,
+                embed_dim=embed_dim)
+            self.num_patches = (
+                (img_size // patch_size)
+                * (img_size // patch_size)
+            )
+
+        # Position embedding
+        self.uniform_power = uniform_power
+        self.pos_embed = None
+        self.pos_embed = nn.Parameter(
+            torch.zeros(1, self.num_patches, embed_dim),
+            requires_grad=False)
+
+        # Attention Blocks
+        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,
+                act_layer=nn.GELU,
+                grid_size=grid_size,
+                grid_depth=grid_depth,
+                attn_drop=attn_drop_rate,
+                norm_layer=norm_layer)
+            for i in range(depth)])
+        self.norm = norm_layer(embed_dim)
+
+        # ------ initialize weights
+        if self.pos_embed is not None:
+            self._init_pos_embed(self.pos_embed.data)  # sincos pos-embed
+        self.init_std = init_std
+        self.apply(self._init_weights)
+        self._rescale_blocks()
+
+    def _init_pos_embed(self, pos_embed):
+        embed_dim = pos_embed.size(-1)
+        grid_size = self.input_size // self.patch_size
+        if self.is_video:
+            grid_depth = self.num_frames // self.tubelet_size
+            sincos = get_3d_sincos_pos_embed(
+                embed_dim,
+                grid_size,
+                grid_depth,
+                cls_token=False,
+                uniform_power=self.uniform_power
+            )
+        else:
+            sincos = get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False)
+        pos_embed.copy_(torch.from_numpy(sincos).float().unsqueeze(0))
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=self.init_std)
+            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)
+        elif isinstance(m, nn.Conv2d):
+            trunc_normal_(m.weight, std=self.init_std)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.Conv3d):
+            trunc_normal_(m.weight, std=self.init_std)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+
+    def _rescale_blocks(self):
+        def rescale(param, layer_id):
+            param.div_(math.sqrt(2.0 * layer_id))
+
+        for layer_id, layer in enumerate(self.blocks):
+            rescale(layer.attn.proj.weight.data, layer_id + 1)
+            rescale(layer.mlp.fc2.weight.data, layer_id + 1)
+
+    def get_num_layers(self):
+        return len(self.blocks)
+
+    def no_weight_decay(self):
+        return {}
+
+    def forward(self, x, masks=None):
+        """
+        :param x: input image/video
+        :param masks: indices of patch tokens to mask (remove)
+        """
+        if masks is not None and not isinstance(masks, list):
+            masks = [masks]
+
+        # Tokenize input
+        pos_embed = self.pos_embed
+        if pos_embed is not None:
+            pos_embed = self.interpolate_pos_encoding(x, pos_embed)
+        x = self.patch_embed(x)
+        if pos_embed is not None:
+            x += pos_embed
+        B, N, D = x.shape
+
+        # Mask away unwanted tokens (if masks provided)
+        if masks is not None:
+            x = apply_masks(x, masks)
+            masks = torch.cat(masks, dim=0)
+
+        # Fwd prop
+        outs = []
+        for i, blk in enumerate(self.blocks):
+            x = blk(x, mask=masks)
+            if self.out_layers is not None and i in self.out_layers:
+                outs.append(self.norm(x))
+
+        if self.out_layers is not None:
+            return outs
+
+        if self.norm is not None:
+            x = self.norm(x)
+
+        return x
+
+    def interpolate_pos_encoding(self, x, pos_embed):
+
+        _, N, dim = pos_embed.shape
+
+        if self.is_video:
+
+            # If pos_embed already corret size, just return
+            _, _, T, H, W = x.shape
+            if H == self.input_size and W == self.input_size and T == self.num_frames:
+                return pos_embed
+
+            # Convert depth, height, width of input to be measured in patches
+            # instead of pixels/frames
+            T = T // self.tubelet_size
+            H = H // self.patch_size
+            W = W // self.patch_size
+
+            # Compute the initialized shape of the positional embedding measured
+            # in patches
+            N_t = self.num_frames // self.tubelet_size
+            N_h = N_w = self.input_size // self.patch_size
+            assert N_h * N_w * N_t == N, 'Positional embedding initialized incorrectly'
+
+            # Compute scale factor for spatio-temporal interpolation
+            scale_factor = (T/N_t, H/N_h, W/N_w)
+
+            pos_embed = nn.functional.interpolate(
+                pos_embed.reshape(1, N_t, N_h, N_w, dim).permute(0, 4, 1, 2, 3),
+                scale_factor=scale_factor,
+                mode='trilinear')
+            pos_embed = pos_embed.permute(0, 2, 3, 4, 1).view(1, -1, dim)
+            return pos_embed
+
+        else:
+
+            # If pos_embed already corret size, just return
+            _, _, H, W = x.shape
+            if H == self.input_size and W == self.input_size:
+                return pos_embed
+
+            # Compute scale factor for spatial interpolation
+            npatch = (H // self.patch_size) * (W // self.patch_size)
+            scale_factor = math.sqrt(npatch / N)
+
+            pos_embed = nn.functional.interpolate(
+                pos_embed.reshape(1, int(math.sqrt(N)), int(math.sqrt(N)), dim).permute(0, 3, 1, 2),
+                scale_factor=scale_factor,
+                mode='bicubic')
+            pos_embed = pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)
+            return pos_embed
+
+
+def vit_tiny(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=192, depth=12, num_heads=3, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+
+def vit_small(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+
+def vit_base(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+
+def vit_large(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1024, depth=24, num_heads=16, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+def vit_huge(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1280, depth=32, num_heads=16, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+
+def vit_giant(patch_size=16, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1408, depth=40, num_heads=16, mlp_ratio=48/11,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
+    return model
+
+
+def vit_gigantic(patch_size=14, **kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1664, depth=48, num_heads=16, mpl_ratio=64/13,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs
+    )
+    return model
+
+
+VIT_EMBED_DIMS = {
+    'vit_tiny': 192,
+    'vit_small': 384,
+    'vit_base': 768,
+    'vit_large': 1024,
+    'vit_huge': 1280,
+    'vit_giant': 1408,
+    'vit_gigantic': 1664,
+}
+
+################
+### Video Encoders ###
+def vit_large_16(patch_size=16, num_frames=16,**kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1024, depth=16, num_heads=16, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), num_frames=num_frames,**kwargs)
+    return model
+
+def vit_huge_16(patch_size=16, num_frames=16,**kwargs):
+    model = VisionTransformer(
+        patch_size=patch_size, embed_dim=1280, depth=16, num_heads=16, mlp_ratio=4,
+        qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), num_frames=num_frames,**kwargs)
+    return model
+################
+
+if __name__ == '__main__':
+    model = vit_large_16()
+    print('Right.')