MLMF.py

"""
paper: Efficient Low-rank Multimodal Fusion with Modality-Specific Factors
ref: https://github.com/Justin1904/Low-rank-Multimodal-Fusion
"""
from __future__ import print_function
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from torch.nn.parameter import Parameter
from torch.nn.init import xavier_normal_

__all__ = ['MLMF']

class SubNet(nn.Module):
    '''
    The subnetwork that is used in LMF for video and audio in the pre-fusion stage
    '''

    def __init__(self, in_size, hidden_size, dropout):
        '''
        Args:
            in_size: input dimension
            hidden_size: hidden layer dimension
            dropout: dropout probability
        Output:
            (return value in forward) a tensor of shape (batch_size, hidden_size)
        '''
        super(SubNet, self).__init__()
        self.norm = nn.BatchNorm1d(in_size)
        self.drop = nn.Dropout(p=dropout)
        self.linear_1 = nn.Linear(in_size, hidden_size)
        self.linear_2 = nn.Linear(hidden_size, hidden_size)
        self.linear_3 = nn.Linear(hidden_size, hidden_size)

    def forward(self, x):
        '''
        Args:
            x: tensor of shape (batch_size, in_size)
        '''
        normed = self.norm(x)
        dropped = self.drop(normed)
        y_1 = F.relu(self.linear_1(dropped), inplace=True)
        y_2 = F.relu(self.linear_2(y_1), inplace=True)
        y_3 = F.relu(self.linear_3(y_2), inplace=True)

        return y_3


class TextSubNet(nn.Module):
    '''
    The LSTM-based subnetwork that is used in LMF for text
    '''

    def __init__(self, in_size, hidden_size, out_size, num_layers=1, dropout=0.2, bidirectional=False):
        '''
        Args:
            in_size: input dimension
            hidden_size: hidden layer dimension
            num_layers: specify the number of layers of LSTMs.
            dropout: dropout probability
            bidirectional: specify usage of bidirectional LSTM
        Output:
            (return value in forward) a tensor of shape (batch_size, out_size)
        '''
        super(TextSubNet, self).__init__()
        if num_layers == 1:
            dropout = 0.0
        self.rnn = nn.LSTM(in_size, hidden_size, num_layers=num_layers, dropout=dropout, bidirectional=bidirectional, batch_first=True)
        self.dropout = nn.Dropout(dropout)
        self.linear_1 = nn.Linear(hidden_size, out_size)

    def forward(self, x):
        '''
        Args:
            x: tensor of shape (batch_size, sequence_len, in_size)
        '''
        _, final_states = self.rnn(x)
        h = self.dropout(final_states[0].squeeze())
        y_1 = self.linear_1(h)
        return y_1


class MLMF(nn.Module):
    '''
    Multi-task Low-rank Multimodal Fusion
    '''

    def __init__(self, args):
        '''
        Args:
            input_dims - a length-3 tuple, contains (audio_dim, video_dim, text_dim)
            hidden_dims - another length-3 tuple, hidden dims of the sub-networks
            text_out - int, specifying the resulting dimensions of the text subnetwork
            dropouts - a length-4 tuple, contains (audio_dropout, video_dropout, text_dropout, post_fusion_dropout)
            output_dim - int, specifying the size of output
            rank - int, specifying the size of rank in LMF
        Output:
            (return value in forward) a scalar value between -3 and 3
        '''
        super(MLMF, self).__init__()

        # dimensions are specified in the order of audio, video and text
        self.text_in, self.audio_in, self.video_in = args.feature_dims
        self.text_hidden, self.audio_hidden, self.video_hidden = args.hidden_dims

        self.text_out= self.text_hidden // 2
        self.output_dim = args.output_dim 
        self.rank = args.rank
        self.use_softmax = args.use_softmax

        self.audio_prob, self.video_prob, self.text_prob, self.post_fusion_prob = args.dropouts

        self.post_text_prob, self.post_audio_prob, self.post_video_prob = args.post_dropouts
        self.post_text_dim = args.post_text_dim
        self.post_audio_dim = args.post_audio_dim
        self.post_video_dim = args.post_video_dim

        # define the pre-fusion subnetworks
        self.audio_subnet = SubNet(self.audio_in, self.audio_hidden, self.audio_prob)
        self.video_subnet = SubNet(self.video_in, self.video_hidden, self.video_prob)
        self.text_subnet = TextSubNet(self.text_in, self.text_hidden, self.text_out, dropout=self.text_prob)

        # self.post_fusion_layer_1 = nn.Linear((self.text_out + 1) * (self.video_hidden + 1) * (self.audio_hidden + 1), self.post_fusion_dim)
        self.audio_factor = Parameter(torch.Tensor(self.rank, self.audio_hidden + 1, self.output_dim))
        self.video_factor = Parameter(torch.Tensor(self.rank, self.video_hidden + 1, self.output_dim))
        self.text_factor = Parameter(torch.Tensor(self.rank, self.text_out + 1, self.output_dim))

        # define the classify layer for text
        self.post_text_dropout = nn.Dropout(p=self.post_text_prob)
        self.post_text_layer_1 = nn.Linear(self.text_out, self.post_text_dim)
        self.post_text_layer_2 = nn.Linear(self.post_text_dim, self.post_text_dim)
        self.post_text_layer_3 = nn.Linear(self.post_text_dim, 1)

        # define the classify layer for audio
        self.post_audio_dropout = nn.Dropout(p=self.post_audio_prob)
        self.post_audio_layer_1 = nn.Linear(self.audio_hidden, self.post_audio_dim)
        self.post_audio_layer_2 = nn.Linear(self.post_audio_dim, self.post_audio_dim)
        self.post_audio_layer_3 = nn.Linear(self.post_audio_dim, 1)

        # define the classify layer for video
        self.post_video_dropout = nn.Dropout(p=self.post_video_prob)
        self.post_video_layer_1 = nn.Linear(self.video_hidden, self.post_video_dim)
        self.post_video_layer_2 = nn.Linear(self.post_video_dim, self.post_video_dim)
        self.post_video_layer_3 = nn.Linear(self.post_video_dim, 1)

        self.fusion_weights = Parameter(torch.Tensor(1, self.rank))
        self.fusion_bias = Parameter(torch.Tensor(1, self.output_dim))

        # init teh factors
        xavier_normal_(self.audio_factor)
        xavier_normal_(self.video_factor)
        xavier_normal_(self.text_factor)
        xavier_normal_(self.fusion_weights)
        self.fusion_bias.data.fill_(0)

    def forward(self, text_x, audio_x, video_x):
        '''
        Args:
            audio_x: tensor of shape (batch_size, audio_in)
            video_x: tensor of shape (batch_size, video_in)
            text_x: tensor of shape (batch_size, sequence_len, text_in)
        '''
        audio_x = audio_x.squeeze(1)
        video_x = video_x.squeeze(1)

        audio_h = self.audio_subnet(audio_x)
        video_h = self.video_subnet(video_x)
        text_h = self.text_subnet(text_x)

        # text
        x_t = self.post_text_dropout(text_h)
        x_t = F.relu(self.post_text_layer_1(x_t), inplace=True)
        x_t = F.relu(self.post_text_layer_2(x_t), inplace=True)
        output_text = self.post_text_layer_3(x_t)
        # audio
        x_a = self.post_audio_dropout(audio_h)
        x_a = F.relu(self.post_audio_layer_1(x_a), inplace=True)
        x_a = F.relu(self.post_audio_layer_2(x_a), inplace=True)
        output_audio = self.post_audio_layer_3(x_a)
        # video
        x_v = self.post_video_dropout(video_h)
        x_v = F.relu(self.post_video_layer_1(video_h), inplace=True)
        x_v = F.relu(self.post_video_layer_2(x_v), inplace=True)
        output_video = self.post_video_layer_3(x_v)

        batch_size = audio_h.data.shape[0]

        # next we perform low-rank multimodal fusion
        # here is a more efficient implementation than the one the paper describes
        # basically swapping the order of summation and elementwise product
        # next we perform "tensor fusion", which is essentially appending 1s to the tensors and take Kronecker product
        add_one = torch.ones(size=[batch_size, 1], requires_grad=False).type_as(audio_h).to(text_x.device)
        _audio_h = torch.cat((add_one, audio_h), dim=1)
        _video_h = torch.cat((add_one, video_h), dim=1)
        _text_h = torch.cat((add_one, text_h), dim=1)
        
        fusion_audio = torch.matmul(_audio_h, self.audio_factor)
        fusion_video = torch.matmul(_video_h, self.video_factor)
        fusion_text = torch.matmul(_text_h, self.text_factor)

        # fusion
        fusion_zy = fusion_audio * fusion_video * fusion_text

        # output = torch.sum(fusion_zy, dim=0).squeeze()
        # use linear transformation instead of simple summation, more flexibility
        output = torch.matmul(self.fusion_weights, fusion_zy.permute(1, 0, 2)).squeeze() + self.fusion_bias
        output = output.view(-1, self.output_dim)
        if self.use_softmax:
            output = F.softmax(output)
        res = {
            'Feature_t': text_h,
            'Feature_a': audio_h,
            'Feature_v': video_h,
            'Feature_f': fusion_zy.permute(1, 0, 2).squeeze(),
            'M': output,
            'T': output_text,
            'A': output_audio,
            'V': output_video
        }
        return res