modeling_videomae.py

# coding=utf-8
# Copyright 2022 Multimedia Computing Group, Nanjing University and The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" PyTorch VideoMAE (masked autoencoder) model."""


import collections.abc
import math
from copy import deepcopy
from dataclasses import dataclass
from typing import Optional, Set, Tuple, Union

import numpy as np
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss

from ...activations import ACT2FN
from ...image_utils import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from ...modeling_outputs import BaseModelOutput, ImageClassifierOutput
from ...modeling_utils import PreTrainedModel
from ...pytorch_utils import find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import (
    ModelOutput,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    logging,
    replace_return_docstrings,
)
from .configuration_videomae import VideoMAEConfig


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "VideoMAEConfig"
_CHECKPOINT_FOR_DOC = "nanjing/videomae-base"

VIDEOMAE_PRETRAINED_MODEL_ARCHIVE_LIST = [
    # TODO rename to organization
    "nielsr/videomae-base",
    # See all VideoMAE models at https://huggingface.co/models?filter=videomae
]


@dataclass
class VideoMAEDecoderOutput(ModelOutput):
    """
    Class for VideoMAEDecoder's outputs, with potential hidden states and attentions.

    Args:
        logits (`torch.FloatTensor` of shape `(batch_size, patch_size ** 2 * num_channels)`):
            Pixel reconstruction logits.
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer
            plus the initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in
            the self-attention heads.
    """

    logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


@dataclass
class VideoMAEForPreTrainingOutput(ModelOutput):
    """
    Class for VideoMAEForPreTraining's outputs, with potential hidden states and attentions.

    Args:
        loss (`torch.FloatTensor` of shape `(1,)`):
            Pixel reconstruction loss.
        logits (`torch.FloatTensor` of shape `(batch_size, patch_size ** 2 * num_channels)`):
            Pixel reconstruction logits.
        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
            shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer
            plus the initial embedding outputs.
        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
            sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in
            the self-attention heads.
    """

    loss: Optional[torch.FloatTensor] = None
    logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


# sin-cos position encoding
# https://github.com/jadore801120/attention-is-all-you-need-pytorch/blob/master/transformer/Models.py#L31
def get_sinusoid_encoding_table(n_position, d_hid):
    """Sinusoid position encoding table"""
    # TODO: make it with torch instead of numpy
    def get_position_angle_vec(position):
        return [position / np.power(10000, 2 * (hid_j // 2) / d_hid) for hid_j in range(d_hid)]

    sinusoid_table = np.array([get_position_angle_vec(pos_i) for pos_i in range(n_position)])
    sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2])  # dim 2i
    sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2])  # dim 2i+1

    return torch.FloatTensor(sinusoid_table).unsqueeze(0)


class VideoMAEEmbeddings(nn.Module):
    """
    Construct the patch and position embeddings.

    """

    def __init__(self, config):
        super().__init__()

        self.patch_embeddings = VideoMAEPatchEmbeddings(config)
        self.num_patches = self.patch_embeddings.num_patches
        # fixed sin-cos embedding
        self.position_embeddings = get_sinusoid_encoding_table(self.num_patches, config.hidden_size)
        self.config = config

    def forward(self, pixel_values, bool_masked_pos):
        # create patch embeddings
        embeddings = self.patch_embeddings(pixel_values)

        # add position embeddings
        embeddings = embeddings + self.position_embeddings.type_as(embeddings).to(embeddings.device).clone().detach()

        # only keep visible patches
        # ~bool_masked_pos means visible
        if bool_masked_pos is not None:
            batch_size, _, num_channels = embeddings.shape
            embeddings = embeddings[~bool_masked_pos]
            embeddings = embeddings.reshape(batch_size, -1, num_channels)

        return embeddings


class VideoMAEPatchEmbeddings(nn.Module):
    """
    Video to Patch Embedding. This module turns a batch of videos of shape (batch_size, num_frames, num_channels,
    height, width) into a tensor of shape (batch_size, seq_len, hidden_size) to be consumed by a Transformer encoder.

    The seq_len (the number of patches) equals (number of frames // tubelet_size) * (height // patch_size) * (width //
    patch_size).

    """

    def __init__(self, config):
        image_size, patch_size, num_channels, hidden_size, num_frames, tubelet_size = (
            config.image_size,
            config.patch_size,
            config.num_channels,
            config.hidden_size,
            config.num_frames,
            config.tubelet_size,
        )

        super().__init__()
        image_size = image_size if isinstance(image_size, collections.abc.Iterable) else (image_size, image_size)
        patch_size = patch_size if isinstance(patch_size, collections.abc.Iterable) else (patch_size, patch_size)
        self.image_size = image_size
        self.patch_size = patch_size
        self.tubelet_size = int(tubelet_size)
        num_patches = (
            (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0]) * (num_frames // self.tubelet_size)
        )
        self.num_channels = num_channels
        self.num_patches = num_patches
        self.projection = nn.Conv3d(
            in_channels=num_channels,
            out_channels=hidden_size,
            kernel_size=(self.tubelet_size, patch_size[0], patch_size[1]),
            stride=(self.tubelet_size, patch_size[0], patch_size[1]),
        )

    def forward(self, pixel_values):
        batch_size, num_frames, num_channels, height, width = pixel_values.shape
        if num_channels != self.num_channels:
            raise ValueError(
                "Make sure that the channel dimension of the pixel values match with the one set in the configuration."
            )
        if height != self.image_size[0] or width != self.image_size[1]:
            raise ValueError(
                f"Input image size ({height}*{width}) doesn't match model ({self.image_size[0]}*{self.image_size[1]})."
            )
        # permute to (batch_size, num_channels, num_frames, height, width)
        pixel_values = pixel_values.permute(0, 2, 1, 3, 4)
        embeddings = self.projection(pixel_values).flatten(2).transpose(1, 2)
        return embeddings


class VideoMAESelfAttention(nn.Module):
    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                f"The hidden size {config.hidden_size,} is not a multiple of the number of attention "
                f"heads {config.num_attention_heads}."
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size, bias=False)
        self.key = nn.Linear(config.hidden_size, self.all_head_size, bias=False)
        self.value = nn.Linear(config.hidden_size, self.all_head_size, bias=False)

        if config.qkv_bias:
            self.q_bias = nn.Parameter(torch.zeros(self.all_head_size))
            self.v_bias = nn.Parameter(torch.zeros(self.all_head_size))
        else:
            self.q_bias = None
            self.v_bias = None

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)

    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(
        self, hidden_states, head_mask: Optional[torch.Tensor] = None, output_attentions: bool = False
    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:

        k_bias = torch.zeros_like(self.v_bias, requires_grad=False) if self.q_bias is not None else None
        keys = nn.functional.linear(input=hidden_states, weight=self.key.weight, bias=k_bias)
        values = nn.functional.linear(input=hidden_states, weight=self.value.weight, bias=self.v_bias)
        queries = nn.functional.linear(input=hidden_states, weight=self.query.weight, bias=self.q_bias)

        key_layer = self.transpose_for_scores(keys)
        value_layer = self.transpose_for_scores(values)
        query_layer = self.transpose_for_scores(queries)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)

        # Normalize the attention scores to probabilities.
        attention_probs = nn.functional.softmax(attention_scores, dim=-1)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs = attention_probs * head_mask

        context_layer = torch.matmul(attention_probs, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(new_context_layer_shape)

        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)

        return outputs


# Copied from transformers.models.vit.modeling_vit.ViTSelfOutput with ViT->VideoMAE
class VideoMAESelfOutput(nn.Module):
    """
    The residual connection is defined in VideoMAELayer instead of here (as is the case with other models), due to the
    layernorm applied before each block.
    """

    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:

        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)

        return hidden_states


# Copied from transformers.models.vit.modeling_vit.ViTAttention with ViT->VideoMAE
class VideoMAEAttention(nn.Module):
    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.attention = VideoMAESelfAttention(config)
        self.output = VideoMAESelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads: Set[int]) -> None:
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.attention.num_attention_heads, self.attention.attention_head_size, self.pruned_heads
        )

        # Prune linear layers
        self.attention.query = prune_linear_layer(self.attention.query, index)
        self.attention.key = prune_linear_layer(self.attention.key, index)
        self.attention.value = prune_linear_layer(self.attention.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.attention.num_attention_heads = self.attention.num_attention_heads - len(heads)
        self.attention.all_head_size = self.attention.attention_head_size * self.attention.num_attention_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    def forward(
        self,
        hidden_states: torch.Tensor,
        head_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
        self_outputs = self.attention(hidden_states, head_mask, output_attentions)

        attention_output = self.output(self_outputs[0], hidden_states)

        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs


# Copied from transformers.models.vit.modeling_vit.ViTIntermediate ViT->VideoMAE
class VideoMAEIntermediate(nn.Module):
    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:

        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)

        return hidden_states


# Copied from transformers.models.vit.modeling_vit.ViTOutput ViT->VideoMAE
class VideoMAEOutput(nn.Module):
    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)

        hidden_states = hidden_states + input_tensor

        return hidden_states


# Copied from transformers.models.vit.modeling_vit.ViTLayer with ViT->VideoMAE
class VideoMAELayer(nn.Module):
    """This corresponds to the Block class in the timm implementation."""

    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = VideoMAEAttention(config)
        self.intermediate = VideoMAEIntermediate(config)
        self.output = VideoMAEOutput(config)
        self.layernorm_before = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.layernorm_after = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        head_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
    ) -> Union[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor]]:
        self_attention_outputs = self.attention(
            self.layernorm_before(hidden_states),  # in VideoMAE, layernorm is applied before self-attention
            head_mask,
            output_attentions=output_attentions,
        )
        attention_output = self_attention_outputs[0]
        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights

        # first residual connection
        hidden_states = attention_output + hidden_states

        # in VideoMAE, layernorm is also applied after self-attention
        layer_output = self.layernorm_after(hidden_states)
        layer_output = self.intermediate(layer_output)

        # second residual connection is done here
        layer_output = self.output(layer_output, hidden_states)

        outputs = (layer_output,) + outputs

        return outputs


# Copied from transformers.models.vit.modeling_vit.ViTEncoder with ViT->VideoMAE
class VideoMAEEncoder(nn.Module):
    def __init__(self, config: VideoMAEConfig) -> None:
        super().__init__()
        self.config = config
        self.layer = nn.ModuleList([VideoMAELayer(config) for _ in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states: torch.Tensor,
        head_mask: Optional[torch.Tensor] = None,
        output_attentions: bool = False,
        output_hidden_states: bool = False,
        return_dict: bool = True,
    ) -> Union[tuple, BaseModelOutput]:
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None

        for i, layer_module in enumerate(self.layer):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, output_attentions)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    layer_head_mask,
                )
            else:
                layer_outputs = layer_module(hidden_states, layer_head_mask, output_attentions)

            hidden_states = layer_outputs[0]

            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, all_hidden_states, all_self_attentions] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


class VideoMAEPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = VideoMAEConfig
    base_model_prefix = "videomae"
    main_input_name = "pixel_values"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        if isinstance(module, (nn.Linear, nn.Conv3d)):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, VideoMAEEncoder):
            module.gradient_checkpointing = value


VIDEOMAE_START_DOCSTRING = r"""
    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it
    as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
    behavior.

    Parameters:
        config ([`VideoMAEConfig`]): Model configuration class with all the parameters of the model.
            Initializing with a config file does not load the weights associated with the model, only the
            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""

VIDEOMAE_INPUTS_DOCSTRING = r"""
    Args:
        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_frames, num_channels, height, width)`):
            Pixel values. Pixel values can be obtained using [`VideoMAEFeatureExtractor`]. See
            [`VideoMAEFeatureExtractor.__call__`] for details.

        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:

            - 1 indicates the head is **not masked**,
            - 0 indicates the head is **masked**.

        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
"""


@add_start_docstrings(
    "The bare VideoMAE Model transformer outputting raw hidden-states without any specific head on top.",
    VIDEOMAE_START_DOCSTRING,
)
class VideoMAEModel(VideoMAEPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config

        self.embeddings = VideoMAEEmbeddings(config)
        self.encoder = VideoMAEEncoder(config)

        self.layernorm = (
            nn.Identity() if config.use_mean_pooling else nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        )

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embeddings.patch_embeddings

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
        class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    @add_start_docstrings_to_model_forward(VIDEOMAE_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=BaseModelOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        pixel_values,
        bool_masked_pos=None,
        head_mask=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        Returns:

        Examples:

        ```python
        >>> from transformers import VideoMAEFeatureExtractor, VideoMAEModel
        >>> import torch

        >>> video = torch.randn(1, 16, 3, 224, 224)

        >>> feature_extractor = VideoMAEFeatureExtractor.from_pretrained("nanjing/videomae-base")
        >>> model = VideoMAEModel.from_pretrained("nanjing/videomae-base")

        >>> inputs = feature_extractor(video, return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> last_hidden_states = outputs.last_hidden_state
        ```"""
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)

        embedding_output = self.embeddings(pixel_values, bool_masked_pos)

        encoder_outputs = self.encoder(
            embedding_output,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        sequence_output = encoder_outputs[0]
        sequence_output = self.layernorm(sequence_output)

        if not return_dict:
            return (sequence_output,) + encoder_outputs[1:]

        return BaseModelOutput(
            last_hidden_state=sequence_output,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


class VideoMAEDecoder(nn.Module):
    def __init__(self, config, num_patches):
        super().__init__()

        decoder_num_labels = config.num_channels * config.tubelet_size * config.patch_size**2

        decoder_config = deepcopy(config)
        decoder_config.hidden_size = config.decoder_hidden_size
        decoder_config.num_hidden_layers = config.decoder_num_hidden_layers
        decoder_config.num_attention_heads = config.decoder_num_attention_heads
        decoder_config.intermediate_size = config.decoder_intermediate_size
        self.decoder_layers = nn.ModuleList(
            [VideoMAELayer(decoder_config) for _ in range(config.decoder_num_hidden_layers)]
        )

        self.decoder_norm = nn.LayerNorm(config.decoder_hidden_size)
        self.decoder_pred = (
            nn.Linear(config.decoder_hidden_size, decoder_num_labels) if decoder_num_labels > 0 else nn.Identity()
        )

        self.gradient_checkpointing = False
        self.config = config

    def forward(
        self,
        hidden_states,
        return_token_num,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
    ):
        # apply Transformer layers (blocks)
        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        for i, layer_module in enumerate(self.decoder_layers):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        return module(*inputs, output_attentions)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer_module),
                    hidden_states,
                    None,
                )
            else:
                layer_outputs = layer_module(hidden_states, head_mask=None, output_attentions=output_attentions)

            hidden_states = layer_outputs[0]

            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if return_token_num > 0:
            hidden_states = hidden_states[:, -return_token_num:]

        # predictor projection
        hidden_states = self.decoder_norm(hidden_states)
        logits = self.decoder_pred(hidden_states)

        if not return_dict:
            return tuple(v for v in [logits, all_hidden_states, all_self_attentions] if v is not None)
        return VideoMAEDecoderOutput(
            logits=logits,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


@add_start_docstrings(
    "The VideoMAE Model transformer with the decoder on top for self-supervised pre-training.",
    VIDEOMAE_START_DOCSTRING,
)
class VideoMAEForPreTraining(VideoMAEPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config

        self.videomae = VideoMAEModel(config)

        self.encoder_to_decoder = nn.Linear(config.hidden_size, config.decoder_hidden_size, bias=False)
        self.mask_token = nn.Parameter(torch.zeros(1, 1, config.decoder_hidden_size))
        self.position_embeddings = get_sinusoid_encoding_table(
            self.videomae.embeddings.num_patches, config.decoder_hidden_size
        )

        self.decoder = VideoMAEDecoder(config, num_patches=self.videomae.embeddings.num_patches)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(VIDEOMAE_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=VideoMAEForPreTrainingOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        pixel_values,
        bool_masked_pos,
        head_mask=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        Returns:

        Examples:
        ```python
        >>> from transformers import VideoMAEFeatureExtractor, VideoMAEForPreTraining
        >>> import numpy as np

        >>> video = np.random.randn(16, 3, 224, 224).tolist()

        >>> feature_extractor = VideoMAEFeatureExtractor.from_pretrained("nanjing/vit-mae-base")
        >>> model = VideoMAEForPreTraining.from_pretrained("nanjing/vit-mae-base")

        >>> pixel_values = feature_extractor(video, return_tensors="pt").pixel_values
        >>> bool_masked_pos = ...

        >>> outputs = model(pixel_values, bool_masked_pos=bool_masked_pos)
        >>> loss = outputs.loss
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.videomae(
            pixel_values,
            bool_masked_pos=bool_masked_pos,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]
        sequence_output = self.encoder_to_decoder(
            sequence_output
        )  # [batch_size, num_visible_patches, decoder_hidden_size]
        batch_size, seq_len, num_channels = sequence_output.shape

        # we don't unshuffle the correct visible token order, but shuffle the position embeddings accordingly.
        if bool_masked_pos is None:
            raise ValueError("One must provided a boolean mask ")
        expanded_position_embeddings = self.position_embeddings.expand(batch_size, -1, -1).type_as(pixel_values)
        expanded_position_embeddings = expanded_position_embeddings.to(pixel_values.device).clone().detach()
        pos_emb_visible = expanded_position_embeddings[~bool_masked_pos].reshape(batch_size, -1, num_channels)
        pos_emb_mask = expanded_position_embeddings[bool_masked_pos].reshape(batch_size, -1, num_channels)

        x_full = torch.cat(
            [sequence_output + pos_emb_visible, self.mask_token + pos_emb_mask], dim=1
        )  # [batch_size, num_patches, decoder_hidden_size]

        decoder_outputs = self.decoder(
            x_full, pos_emb_mask.shape[1]
        )  # [batch_size, num_masked_patches, num_channels * patch_size * patch_size]
        logits = decoder_outputs.logits

        # TODO verify loss computation
        loss = None
        with torch.no_grad():
            # calculate the labels to be predicted
            # first, unnormalize the frames
            device = pixel_values.device
            mean = torch.as_tensor(IMAGENET_DEFAULT_MEAN).to(device)[None, None, :, None, None]
            std = torch.as_tensor(IMAGENET_DEFAULT_STD).to(device)[None, None, :, None, None]
            frames = pixel_values * std + mean  # in [0, 1]

            batch_size, time, num_channels, height, width = frames.shape
            tubelet_size, patch_size = self.config.tubelet_size, self.config.patch_size
            if self.config.norm_pix_loss:
                # step 1: split up dimensions (time by tubelet_size, height by patch_size, width by patch_size)
                frames = frames.view(
                    batch_size,
                    time // tubelet_size,
                    tubelet_size,
                    num_channels,
                    height // patch_size,
                    patch_size,
                    width // patch_size,
                    patch_size,
                )
                # step 2: move dimensions to concatenate: (batch_size, T//ts, H//ps, W//ps, ts, ps, ps, C)
                frames = frames.permute(0, 1, 4, 6, 2, 5, 7, 3).contiguous()
                # step 3: concatenate: (batch_size, T//ts, H//bs, W//bs, ts*bs*bs, C)
                frames = frames.view(
                    batch_size,
                    time // tubelet_size * height // patch_size * width // patch_size,
                    tubelet_size * patch_size * patch_size,
                    num_channels,
                )
                # step 4: normalize. The authors find that the mean is about 0.48 and standard deviation is about 0.08.
                frames_norm = (frames - frames.mean(dim=-2, keepdim=True)) / (
                    frames.var(dim=-2, unbiased=True, keepdim=True).sqrt() + 1e-6
                )
                # step 5: reshape to (batch_size, T//ts * H//ps * W//ps, ts * ps * ps * C)
                videos_patch = frames_norm.view(
                    batch_size,
                    time // tubelet_size * height // patch_size * width // patch_size,
                    tubelet_size * patch_size * patch_size * num_channels,
                )
            else:
                # step 1: split up dimensions (time by tubelet_size, height by patch_size, width by patch_size)
                frames = frames.view(
                    batch_size,
                    time // tubelet_size,
                    tubelet_size,
                    num_channels,
                    height // patch_size,
                    patch_size,
                    width // patch_size,
                    patch_size,
                )
                # step 2: move dimensions to concatenate: (batch_size, T//ts, H//ps, W//ps, ts, ps, ps, C)
                frames = frames.permute(0, 1, 4, 6, 2, 5, 7, 3).contiguous()
                # step 3: concatenate
                videos_patch = frames.view(
                    batch_size,
                    time // tubelet_size * height // patch_size * width // patch_size,
                    tubelet_size * patch_size * patch_size * num_channels,
                )

            batch_size, _, num_channels = videos_patch.shape
            labels = videos_patch[bool_masked_pos].reshape(batch_size, -1, num_channels)

        loss_fct = MSELoss()
        loss = loss_fct(logits, labels)

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return VideoMAEForPreTrainingOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


@add_start_docstrings(
    """VideoMAE Model transformer with a video classification head on top (a linear layer on top of the final hidden state of
    the [CLS] token) e.g. for ImageNet.""",
    VIDEOMAE_START_DOCSTRING,
)
class VideoMAEForVideoClassification(VideoMAEPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.num_labels = config.num_labels
        self.videomae = VideoMAEModel(config)

        # Classifier head
        self.fc_norm = nn.LayerNorm(config.hidden_size) if config.use_mean_pooling else None
        self.classifier = nn.Linear(config.hidden_size, config.num_labels) if config.num_labels > 0 else nn.Identity()

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(VIDEOMAE_INPUTS_DOCSTRING)
    @replace_return_docstrings(output_type=ImageClassifierOutput, config_class=_CONFIG_FOR_DOC)
    def forward(
        self,
        pixel_values: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ):
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).

        Returns:

        Examples:

        ```python
        >>> from transformers import VideoMAEFeatureExtractor, VideoMAEForVideoClassification
        >>> import numpy as np

        >>> video = np.random.randn(16, 3, 224, 224).tolist()

        >>> feature_extractor = VideoMAEFeatureExtractor.from_pretrained("nanjing/videomae-base")
        >>> model = VideoMAEForVideoClassification.from_pretrained("nanjing/videomae-base")

        >>> inputs = feature_extractor(video, return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> last_hidden_states = outputs.last_hidden_state
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.videomae(
            pixel_values,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        if self.fc_norm is not None:
            sequence_output = self.fc_norm(sequence_output.mean(1))
        else:
            sequence_output = sequence_output[:, 0]

        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    self.config.problem_type = "regression"
                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
                    self.config.problem_type = "single_label_classification"
                else:
                    self.config.problem_type = "multi_label_classification"

            if self.config.problem_type == "regression":
                loss_fct = MSELoss()
                if self.num_labels == 1:
                    loss = loss_fct(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = CrossEntropyLoss()
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)

        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return ImageClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )