models3.py

import torch
from torch import nn
import torchvision
from torch.nn import functional as F

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # cuda:0


class CNN_Encoder(nn.Module):
    """
    CNN_Encoder.
    """

    def __init__(self, NetType='resnet50', encoded_image_size=14, attention_method="ByPixel"):
        super(CNN_Encoder, self).__init__()
        self.enc_image_size = encoded_image_size
        self.attention_method = attention_method

        # resnet = torchvision.models.resnet101(pretrained=True)  # pretrained ImageNet ResNet-101
        net = torchvision.models.inception_v3(pretrained=True, transform_input=False) if NetType == 'inception_v3' else \
              torchvision.models.vgg16(pretrained=True) if NetType == 'vgg16' else \
              torchvision.models.resnet50(pretrained=True) if NetType == 'resnet50' else torchvision.models.resnet50(pretrained=True)
        # Remove linear and pool layers (since we're not doing classification)
        # Specifically, Remove: AdaptiveAvgPool2d(output_size=(1, 1)), Linear(in_features=2048, out_features=1000, bias=True)]

        # modules = list(net.children())[:-2]
        modules = list(net.children())[:-1] if NetType == 'inception_v3' or NetType == 'vgg16' else list(net.children())[:-2]
        # modules = list(net.children())[:-1] if NetType == 'inception_v3' else list(net.children())[:-2]  # -2 for resnet & vgg
        if NetType == 'inception_v3': del modules[13]

        self.net = nn.Sequential(*modules)

        # every block of resnet for fusion
        if NetType == 'resnet50' or NetType == 'resnet101' or NetType == 'resnet152':
            resnet_block1 = list(net.children())[:5]
            self.resnet_block1 = nn.Sequential(*resnet_block1)
            resnet_block2 = list(net.children())[5]
            self.resnet_block2 = nn.Sequential(*resnet_block2)
            resnet_block3 = list(net.children())[6]
            self.resnet_block3 = nn.Sequential(*resnet_block3)
            resnet_block4 = list(net.children())[7]
            self.resnet_block4 = nn.Sequential(*resnet_block4)
            self.conv4 = nn.Conv2d(in_channels=2048, out_channels=1024, kernel_size=1, stride=1)
            self.conv3 = nn.Conv2d(in_channels=1024, out_channels=512, kernel_size=1, stride=1)
            self.conv2 = nn.Conv2d(in_channels=512, out_channels=512, kernel_size=1, stride=1)



        if self.attention_method == "ByChannel":
            self.cnn1 = nn.Conv2d(in_channels=2048, out_channels=512, kernel_size=(1, 1), stride=(1, 1), bias=False)
            self.bn1 = nn.BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
            self.relu = nn.ReLU(inplace=True)

        # Resize image to fixed size to allow input images of variable size
        self.adaptive_pool = nn.AdaptiveAvgPool2d((encoded_image_size, encoded_image_size))
        self.adaptive_pool4 = nn.AdaptiveAvgPool2d((encoded_image_size, encoded_image_size))
        self.adaptive_pool3 = nn.AdaptiveAvgPool2d((encoded_image_size, encoded_image_size))

        self.fine_tune()

    def forward(self, images):
        """
        Forward propagation.

        :param images: images, a tensor of dimensions (batch_size, 3, image_size, image_size)
        :return: encoded images [batch_size, encoded_image_size=14, encoded_image_size=14, 2048]
        """


        # with fusion for resnet
        out1 = self.resnet_block1(images)  # 256
        out2 = self.resnet_block2(out1)  # 512
        out3 = self.resnet_block3(out2)  # 1024
        out4 = self.resnet_block4(out3)  # 2048

        # # FIXME:concat43
        # p4 = self.conv4(out4)  # 1536
        # p3 = self.conv3(out3)  # 512
        # out = torch.cat([F.interpolate(p4, scale_factor=2), p3], dim=1)
        # # FIXME:concat432
        p4 = self.conv4(out4)  # 1024
        p3 = self.conv3(out3)  # 512
        p2 = self.conv2(out2)  # 512
        out = torch.cat([F.interpolate(p4, scale_factor=2), p3, F.interpolate(p2, scale_factor=0.5)], dim=1)




        # without fusion
        # out = self.net(images)  # (batch_size, 2048, image_size/32, image_size/32)


        if self.attention_method == "ByChannel":  # [batch_size, 2048, 8, 8] -> # [batch_size, 512, 8, 8]
            out = self.relu(self.bn1(self.cnn1(out)))
        out = self.adaptive_pool(out)  # [batch_size, 2048/512, 8, 8] -> [batch_size, 2048/512, 14, 14] #FIXME:注释 for fusion
        out = out.permute(0, 2, 3, 1)
        return out

        # fusion4 = fusion4.permute(0, 2, 3, 1)
        # fusion3 = fusion3.permute(0, 2, 3, 1)
        # out = [fusion4,fusion3]
        # return out

    def fine_tune(self, fine_tune=True):
        """
        Allow or prevent the computation of gradients for convolutional blocks 2 through 4 of the encoder.

        :param fine_tune: Allow?
        """
        for p in self.net.parameters():
            p.requires_grad = False
        # If fine-tuning, only fine-tune convolutional blocks 2 through 4
        for c in list(self.net.children())[5:]:  # FIXME:maybe try 6:
            for p in c.parameters():
                p.requires_grad = fine_tune


class DecoderLSTM(nn.Module):
    """
    Decoder.
    """
    def __init__(self, embed_dim, decoder_dim, vocab_size, encoder_dim=2048, dropout=0.5):
        """
        :param embed_dim: embedding size
        :param decoder_dim: size of decoder's RNN
        :param vocab_size: size of vocabulary
        :param encoder_dim: feature size of encoded images
        :param dropout: dropout
        """
        super(DecoderLSTM, self).__init__()

        self.encoder_dim = encoder_dim
        self.embed_dim = embed_dim
        self.decoder_dim = decoder_dim
        self.vocab_size = vocab_size
        self.dropout = dropout

        self.embedding = nn.Embedding(vocab_size, embed_dim)  # embedding layer
        self.dropout = nn.Dropout(p=self.dropout)
        # FIXME：feature map输入在每一时刻
        # self.decode_step = nn.LSTMCell(embed_dim + encoder_dim, decoder_dim, bias=True)  # decoding LSTMCell,每一时刻都输入特征图
        # FIXME：feature map输入仅在初始输入
        self.decode_step = nn.LSTMCell(embed_dim, decoder_dim, bias=True)  # decoding LSTMCell，只在初始时刻输入特征

        self.init_h = nn.Linear(encoder_dim, decoder_dim)  # linear layer to find initial hidden state of LSTMCell
        self.init_c = nn.Linear(encoder_dim, decoder_dim)  # linear layer to find initial cell state of LSTMCell
        self.f_beta = nn.Linear(decoder_dim, encoder_dim)  # linear layer to create a sigmoid-activated gate
        self.sigmoid = nn.Sigmoid()
        self.fc = nn.Linear(decoder_dim, vocab_size)  # linear layer to find scores over vocabulary
        self.init_weights()  # initialize some layers with the uniform distribution

    def init_weights(self):
        """
        Initializes some parameters with values from the uniform distribution, for easier convergence.
        """
        self.embedding.weight.data.uniform_(-0.1, 0.1)
        self.fc.bias.data.fill_(0)
        self.fc.weight.data.uniform_(-0.1, 0.1)

    def load_pretrained_embeddings(self, embeddings):
        """
        Loads embedding layer with pre-trained embeddings.

        :param embeddings: pre-trained embeddings
        """
        self.embedding.weight = nn.Parameter(embeddings)

    def fine_tune_embeddings(self, fine_tune=True):
        """
        Allow fine-tuning of embedding layer? (Only makes sense to not-allow if using pre-trained embeddings).

        :param fine_tune: Allow?
        """
        for p in self.embedding.parameters():
            p.requires_grad = fine_tune

    def init_hidden_state(self, encoder_out):
        """
        Creates the initial hidden and cell states for the decoder's LSTM based on the encoded images.

        :param encoder_out: encoded images, a tensor of dimension (batch_size, num_pixels, encoder_dim)
        :return: hidden state, cell state
        """
        mean_encoder_out = encoder_out.mean(dim=1)
        h = self.init_h(mean_encoder_out)  # (batch_size, decoder_dim)
        c = self.init_c(mean_encoder_out)
        return h, c

    def forward(self, encoder_out, encoded_captions, caption_lengths):
        """
        Forward propagation.

        :param encoder_out: encoded images, a tensor of dimension (batch_size, enc_image_size, enc_image_size, encoder_dim)
        :param encoded_captions: encoded captions, a tensor of dimension (batch_size, max_caption_length)
        :param caption_lengths: caption lengths, a tensor of dimension (batch_size, 1)
        :return: scores for vocabulary, sorted encoded captions, decode lengths, weights, sort indices
        """

        batch_size = encoder_out.size(0)
        encoder_dim = encoder_out.size(-1)
        vocab_size = self.vocab_size

        # Flatten image
        encoder_out = encoder_out.view(batch_size, -1, encoder_dim)  # (batch_size, num_pixels, encoder_dim)
        num_pixels = encoder_out.size(1)

        # Sort input data by decreasing lengths; why? apparent below
        caption_lengths, sort_ind = caption_lengths.squeeze(1).sort(dim=0, descending=True)
        encoder_out = encoder_out[sort_ind]
        encoded_captions = encoded_captions[sort_ind]

        # Embedding
        embeddings = self.embedding(encoded_captions)  # (batch_size, max_caption_length, embed_dim)

        # Initialize LSTM state
        h, c = self.init_hidden_state(encoder_out)  # (batch_size, decoder_dim)

        # We won't decode at the <end> position, since we've finished generating as soon as we generate <end>
        # So, decoding lengths are actual lengths - 1
        decode_lengths = (caption_lengths - 1).tolist()

        # Create tensors to hold word predicion scores and alphas
        predictions = torch.zeros(batch_size, max(decode_lengths), vocab_size).to(device)

        # At each time-step, decode by
        # attention-weighing the encoder's output based on the decoder's previous hidden state output
        # then generate a new word in the decoder with the previous word and the attention weighted encoding
        for t in range(max(decode_lengths)):
            batch_size_t = sum([l > t for l in decode_lengths])

            # FIXME：在每一时刻输入feature map
            # h, c = self.decode_step(
            #     torch.cat([embeddings[:batch_size_t, t, :], encoder_out[:batch_size_t].mean(dim=1)], dim=1),
            #     (h[:batch_size_t], c[:batch_size_t]))  # (batch_size_t, decoder_dim)

            # FIXME：仅在初始时刻输入feature map
            h, c = self.decode_step(embeddings[:batch_size_t, t, :],
                                    (h[:batch_size_t], c[:batch_size_t]))

            preds = self.fc(self.dropout(h))  # (batch_size_t, vocab_size)
            predictions[:batch_size_t, t, :] = preds

        return predictions, encoded_captions, decode_lengths, sort_ind


class Attention(nn.Module):
    """
    Attention Network.
    """

    def __init__(self, encoder_dim, decoder_dim, attention_dim):
        """
        :param encoder_dim: feature size of encoded images
        :param decoder_dim: size of decoder's RNN
        :param attention_dim: size of the attention network
        """
        super(Attention, self).__init__()
        self.encoder_att = nn.Linear(encoder_dim, attention_dim)  # linear layer to transform encoded image
        self.decoder_att = nn.Linear(decoder_dim, attention_dim)  # linear layer to transform decoder's output
        self.full_att = nn.Linear(attention_dim, 1)  # linear layer to calculate values to be softmax-ed
        self.relu = nn.ReLU()
        self.softmax = nn.Softmax(dim=1)  # softmax layer to calculate weights

    def forward(self, encoder_out, decoder_hidden):
        """
        Forward propagation.

        :param encoder_out: encoded images, a tensor of dimension (batch_size, num_pixels, encoder_dim)
        :param decoder_hidden: previous decoder output, a tensor of dimension (batch_size, decoder_dim)
        :return: attention weighted encoding, weights
        """
        att1 = self.encoder_att(encoder_out)  # [batch_size_t, num_pixels=196, 2048] -> [batch_size_t, num_pixels, attention_dim]
        att2 = self.decoder_att(decoder_hidden)  # [batch_size_t, decoder_dim=512] -> [batch_size_t, attention_dim]
        att = self.full_att(self.relu(att1 + att2.unsqueeze(1))).squeeze(2)  # [batch_size_t, num_pixels=196, attention_dim] -> [batch_size_t, num_pixels]
        alpha = self.softmax(att)  # [batch_size_t, num_pixels=196]
        attention_weighted_encoding = (encoder_out * alpha.unsqueeze(2)).sum(dim=1)  # [batch_size_t, encoder_dim=2048]

        return attention_weighted_encoding, alpha


class DecoderWithAttention(nn.Module):
    """
    Decoder.
    """

    def __init__(self, attention_dim, embed_dim, decoder_dim, vocab_size, encoder_dim=2048, dropout=0.5):
        """
        :param attention_dim: size of attention network
        :param embed_dim: embedding size
        :param decoder_dim: size of decoder's RNN
        :param vocab_size: size of vocabulary
        :param encoder_dim: feature size of encoded images
        :param dropout: dropout
        """
        super(DecoderWithAttention, self).__init__()

        self.encoder_dim = encoder_dim
        self.attention_dim = attention_dim
        self.embed_dim = embed_dim
        self.decoder_dim = decoder_dim
        self.vocab_size = vocab_size
        self.dropout = dropout

        self.attention = Attention(encoder_dim, decoder_dim, attention_dim)  # attention network

        self.embedding = nn.Embedding(vocab_size, embed_dim, padding_idx=0)  # embedding layer
        self.dropout = nn.Dropout(p=self.dropout)
        self.lstm = nn.LSTMCell(embed_dim + encoder_dim, decoder_dim, bias=True)  # decoding LSTMCell
        self.init_h = nn.Linear(encoder_dim, decoder_dim)  # linear layer to find initial hidden state of LSTMCell
        self.init_c = nn.Linear(encoder_dim, decoder_dim)  # linear layer to find initial cell state of LSTMCell
        self.f_beta = nn.Linear(decoder_dim, encoder_dim)  # linear layer to create a sigmoid-activated gate
        self.sigmoid = nn.Sigmoid()
        self.fc = nn.Linear(decoder_dim, vocab_size)  # linear layer to find scores over vocabulary
        self.init_weights()  # initialize some layers with the uniform distribution
        self.fine_tune_embeddings()

    def init_weights(self):
        """
        Initializes some parameters with values from the uniform distribution, for easier convergence.
        """
        self.embedding.weight.data.uniform_(-0.1, 0.1)
        self.fc.bias.data.fill_(0)
        self.fc.weight.data.uniform_(-0.1, 0.1)

    def load_pretrained_embeddings(self, embeddings):
        """
        Loads embedding layer with pre-trained embeddings.

        :param embeddings: pre-trained embeddings
        """
        self.embedding.weight = nn.Parameter(embeddings)

    def fine_tune_embeddings(self, fine_tune=True):
        """
        Allow fine-tuning of embedding layer? (Only makes sense to not-allow if using pre-trained embeddings).

        :param fine_tune: Allow?
        """
        for p in self.embedding.parameters():
            p.requires_grad = fine_tune

    def init_hidden_state(self, encoder_out):
        """
        Creates the initial hidden and cell states for the decoder's LSTM based on the encoded images.

        :param encoder_out: encoded images, a tensor of dimension (batch_size, num_pixels, encoder_dim)
        :return: hidden state, cell state
        """
        mean_encoder_out = encoder_out.mean(dim=1)  # [batch_size, 196, 2048] -> [batch_size, 2048]
        h = self.init_h(mean_encoder_out)  # (batch_size, decoder_dim)
        c = self.init_c(mean_encoder_out)
        return h, c

    def forward(self, encoder_out, encoded_captions, caption_lengths):
        """
        Forward propagation.

        :param encoder_out: encoded images, a tensor of dimension (batch_size, enc_image_size, enc_image_size, encoder_dim)
        :param encoded_captions: encoded captions, a tensor of dimension (batch_size, max_caption_length)
        :param caption_lengths: caption lengths, a tensor of dimension (batch_size, 1)
        :return: scores for vocabulary, sorted encoded captions, decode lengths, weights, sort indices
        """
        # [batch_size, 14, 14, 2048]/[batch_size, 196, 2048] -> [batch_size, 196, 2048]
        batch_size = encoder_out.size(0)
        encoder_dim = encoder_out.size(-1)
        vocab_size = self.vocab_size

        # Flatten image -> [batch_size, num_pixels=196, encoder_dim=2048]
        encoder_out = encoder_out.view(batch_size, -1, encoder_dim)
        num_pixels = encoder_out.size(1)

        # Sort input data by decreasing lengths; why? For each of data in the batch, when len(prediction) = len(caption_lengths), Stop.
        caption_lengths, sort_ind = caption_lengths.squeeze(1).sort(dim=0, descending=True)
        encoder_out = encoder_out[sort_ind]
        encoded_captions = encoded_captions[sort_ind]

        # Embedding
        embeddings = self.embedding(encoded_captions)  # [batch_size, max_caption_length=52, embed_dim]

        # Initialize LSTM state
        h, c = self.init_hidden_state(encoder_out)  # [batch_size, decoder_dim]

        # We won't decode at the <end> position, since we've finished generating as soon as we generate <end>
        # So, decoding lengths are actual lengths - 1
        decode_lengths = (caption_lengths - 1).tolist()

        # Create tensors to hold word predicion scores and alphas
        predictions = torch.zeros(batch_size, max(decode_lengths), vocab_size).to(device)
        alphas = torch.zeros(batch_size, max(decode_lengths), num_pixels).to(device)

        # At each time-step, decode by
        # attention-weighing the encoder's output based on the decoder's previous hidden state output
        # then generate a new word in the decoder with the previous word and the attention weighted encoding
        for t in range(max(decode_lengths)):
            batch_size_t = sum([l > t for l in decode_lengths])
            # alpha: [batch_size_t, 196]
            # attention_weighted_encoding: [batch_size_t, 2048]
            attention_weighted_encoding, alpha = self.attention(encoder_out[:batch_size_t], h[:batch_size_t])
            gate = self.sigmoid(self.f_beta(h[:batch_size_t]))  # gating scalar, (batch_size_t, encoder_dim)
            attention_weighted_encoding = gate * attention_weighted_encoding
            # torch.cat([batch_size_t, 500], [batch_size_t, 2048], dim=1) = [batch_size_t, 2548] -> [batch_size_t, 512]
            h, c = self.lstm(
                torch.cat([embeddings[:batch_size_t, t, :], attention_weighted_encoding], dim=1),
                (h[:batch_size_t], c[:batch_size_t]))
            preds = self.fc(self.dropout(h))  # [batch_size_t, vocab_size]
            predictions[:batch_size_t, t, :] = preds
            alphas[:batch_size_t, t, :] = alpha

        return predictions, encoded_captions, decode_lengths, alphas, sort_ind