450 lines
19 KiB
Python
450 lines
19 KiB
Python
# PREVALENT, 2020, weituo.hao@duke.edu
|
|
# Modified in Recurrent VLN-BERT, 2020, Yicong.Hong@anu.edu.au
|
|
|
|
from __future__ import absolute_import, division, print_function, unicode_literals
|
|
|
|
import json
|
|
import logging
|
|
import math
|
|
import os
|
|
import sys
|
|
from io import open
|
|
|
|
import torch
|
|
from torch import nn
|
|
from torch.nn import CrossEntropyLoss, MSELoss
|
|
|
|
from transformers.pytorch_transformers.modeling_bert import BertPreTrainedModel, BertConfig
|
|
import pdb
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
def gelu(x):
|
|
"""Implementation of the gelu activation function.
|
|
For information: OpenAI GPT's gelu is slightly different (and gives slightly different results):
|
|
0.5 * x * (1 + torch.tanh(math.sqrt(2 / math.pi) * (x + 0.044715 * torch.pow(x, 3))))
|
|
Also see https://arxiv.org/abs/1606.08415
|
|
"""
|
|
return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))
|
|
|
|
|
|
def swish(x):
|
|
return x * torch.sigmoid(x)
|
|
|
|
|
|
ACT2FN = {"gelu": gelu, "relu": torch.nn.functional.relu, "swish": swish}
|
|
|
|
|
|
try:
|
|
from apex.normalization.fused_layer_norm import FusedLayerNorm as BertLayerNorm
|
|
except (ImportError, AttributeError) as e:
|
|
logger.info("Better speed can be achieved with apex installed from https://www.github.com/nvidia/apex .")
|
|
BertLayerNorm = torch.nn.LayerNorm
|
|
|
|
class BertEmbeddings(nn.Module):
|
|
"""Construct the embeddings from word, position and token_type embeddings.
|
|
"""
|
|
def __init__(self, config):
|
|
super(BertEmbeddings, self).__init__()
|
|
self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=0)
|
|
self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
|
|
self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
|
|
|
|
# self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
|
|
# any TensorFlow checkpoint file
|
|
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
|
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
|
|
|
def forward(self, input_ids, token_type_ids=None, position_ids=None):
|
|
seq_length = input_ids.size(1)
|
|
if position_ids is None:
|
|
position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device)
|
|
position_ids = position_ids.unsqueeze(0).expand_as(input_ids)
|
|
if token_type_ids is None:
|
|
token_type_ids = torch.zeros_like(input_ids)
|
|
|
|
words_embeddings = self.word_embeddings(input_ids)
|
|
position_embeddings = self.position_embeddings(position_ids)
|
|
token_type_embeddings = self.token_type_embeddings(token_type_ids)
|
|
|
|
embeddings = words_embeddings + position_embeddings + token_type_embeddings
|
|
embeddings = self.LayerNorm(embeddings)
|
|
embeddings = self.dropout(embeddings)
|
|
return embeddings
|
|
|
|
|
|
class BertSelfAttention(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertSelfAttention, self).__init__()
|
|
if config.hidden_size % config.num_attention_heads != 0:
|
|
raise ValueError(
|
|
"The hidden size (%d) is not a multiple of the number of attention "
|
|
"heads (%d)" % (config.hidden_size, config.num_attention_heads))
|
|
self.output_attentions = True
|
|
|
|
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)
|
|
self.key = nn.Linear(config.hidden_size, self.all_head_size)
|
|
self.value = nn.Linear(config.hidden_size, self.all_head_size)
|
|
|
|
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
|
|
|
|
def transpose_for_scores(self, x):
|
|
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, attention_mask, head_mask=None):
|
|
mixed_query_layer = self.query(hidden_states)
|
|
mixed_key_layer = self.key(hidden_states)
|
|
mixed_value_layer = self.value(hidden_states)
|
|
|
|
query_layer = self.transpose_for_scores(mixed_query_layer)
|
|
key_layer = self.transpose_for_scores(mixed_key_layer)
|
|
value_layer = self.transpose_for_scores(mixed_value_layer)
|
|
|
|
# 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)
|
|
# Apply the attention mask is (precomputed for all layers in BertModel forward() function)
|
|
attention_scores = attention_scores + attention_mask
|
|
|
|
# Normalize the attention scores to probabilities.
|
|
attention_probs = nn.Softmax(dim=-1)(attention_scores)
|
|
|
|
# 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_scores) if self.output_attentions else (context_layer,)
|
|
return outputs
|
|
|
|
|
|
class BertSelfOutput(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertSelfOutput, self).__init__()
|
|
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
|
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
|
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
|
|
|
def forward(self, hidden_states, input_tensor):
|
|
hidden_states = self.dense(hidden_states)
|
|
hidden_states = self.dropout(hidden_states)
|
|
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
|
return hidden_states
|
|
|
|
|
|
class BertAttention(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertAttention, self).__init__()
|
|
self.self = BertSelfAttention(config)
|
|
self.output = BertSelfOutput(config)
|
|
|
|
def forward(self, input_tensor, attention_mask, head_mask=None):
|
|
self_outputs = self.self(input_tensor, attention_mask, head_mask)
|
|
attention_output = self.output(self_outputs[0], input_tensor)
|
|
outputs = (attention_output,) + self_outputs[1:] # add attentions if we output them
|
|
return outputs
|
|
|
|
|
|
class BertIntermediate(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertIntermediate, self).__init__()
|
|
self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
|
|
if isinstance(config.hidden_act, str) or (sys.version_info[0] == 2 and isinstance(config.hidden_act, unicode)):
|
|
self.intermediate_act_fn = ACT2FN[config.hidden_act]
|
|
else:
|
|
self.intermediate_act_fn = config.hidden_act
|
|
|
|
def forward(self, hidden_states):
|
|
hidden_states = self.dense(hidden_states)
|
|
hidden_states = self.intermediate_act_fn(hidden_states)
|
|
return hidden_states
|
|
|
|
|
|
class BertOutput(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertOutput, self).__init__()
|
|
self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
|
|
self.LayerNorm = BertLayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
|
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
|
|
|
def forward(self, hidden_states, input_tensor):
|
|
hidden_states = self.dense(hidden_states)
|
|
hidden_states = self.dropout(hidden_states)
|
|
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
|
return hidden_states
|
|
|
|
|
|
class BertLayer(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertLayer, self).__init__()
|
|
self.attention = BertAttention(config)
|
|
self.intermediate = BertIntermediate(config)
|
|
self.output = BertOutput(config)
|
|
|
|
def forward(self, hidden_states, attention_mask, head_mask=None):
|
|
attention_outputs = self.attention(hidden_states, attention_mask, head_mask)
|
|
attention_output = attention_outputs[0]
|
|
intermediate_output = self.intermediate(attention_output)
|
|
layer_output = self.output(intermediate_output, attention_output)
|
|
outputs = (layer_output,) + attention_outputs[1:] # add attentions if we output them
|
|
return outputs
|
|
|
|
|
|
class BertPooler(nn.Module):
|
|
def __init__(self, config):
|
|
super(BertPooler, self).__init__()
|
|
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
|
self.activation = nn.Tanh()
|
|
|
|
def forward(self, hidden_states):
|
|
# We "pool" the model by simply taking the hidden state corresponding
|
|
# to the first token.
|
|
first_token_tensor = hidden_states[:, 0]
|
|
pooled_output = self.dense(first_token_tensor)
|
|
pooled_output = self.activation(pooled_output)
|
|
return pooled_output
|
|
|
|
|
|
class BertXAttention(nn.Module):
|
|
def __init__(self, config, ctx_dim=None):
|
|
super().__init__()
|
|
self.att = BertOutAttention(config, ctx_dim=ctx_dim)
|
|
self.output = BertSelfOutput(config)
|
|
|
|
def forward(self, input_tensor, ctx_tensor, ctx_att_mask=None):
|
|
output, attention_scores = self.att(input_tensor, ctx_tensor, ctx_att_mask)
|
|
attention_output = self.output(output, input_tensor)
|
|
return attention_output, attention_scores
|
|
|
|
|
|
class BertOutAttention(nn.Module):
|
|
def __init__(self, config, ctx_dim=None):
|
|
super().__init__()
|
|
if config.hidden_size % config.num_attention_heads != 0:
|
|
raise ValueError(
|
|
"The hidden size (%d) is not a multiple of the number of attention "
|
|
"heads (%d)" % (config.hidden_size, 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
|
|
|
|
# visual_dim = 2048
|
|
if ctx_dim is None:
|
|
ctx_dim =config.hidden_size
|
|
self.query = nn.Linear(config.hidden_size, self.all_head_size)
|
|
self.key = nn.Linear(ctx_dim, self.all_head_size)
|
|
self.value = nn.Linear(ctx_dim, self.all_head_size)
|
|
|
|
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
|
|
|
|
def transpose_for_scores(self, x):
|
|
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, context, attention_mask=None):
|
|
mixed_query_layer = self.query(hidden_states)
|
|
mixed_key_layer = self.key(context)
|
|
mixed_value_layer = self.value(context)
|
|
|
|
query_layer = self.transpose_for_scores(mixed_query_layer)
|
|
key_layer = self.transpose_for_scores(mixed_key_layer)
|
|
value_layer = self.transpose_for_scores(mixed_value_layer)
|
|
|
|
# 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)
|
|
|
|
# Apply the attention mask is (precomputed for all layers in BertModel forward() function)
|
|
if attention_mask is not None:
|
|
attention_scores = attention_scores + attention_mask
|
|
|
|
# Normalize the attention scores to probabilities.
|
|
attention_probs = nn.Softmax(dim=-1)(attention_scores)
|
|
|
|
# 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)
|
|
|
|
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)
|
|
return context_layer, attention_scores
|
|
|
|
|
|
class LXRTXLayer(nn.Module):
|
|
def __init__(self, config):
|
|
super().__init__()
|
|
self.config = config
|
|
# Lang self-att and FFN layer
|
|
self.lang_self_att = BertAttention(config)
|
|
self.lang_inter = BertIntermediate(config)
|
|
self.lang_output = BertOutput(config)
|
|
# Visn self-att and FFN layer
|
|
self.visn_self_att = BertAttention(config)
|
|
self.visn_inter = BertIntermediate(config)
|
|
self.visn_output = BertOutput(config)
|
|
# The cross attention layer
|
|
self.visual_attention = BertXAttention(config)
|
|
|
|
def cross_att(self, lang_input, lang_attention_mask, visn_input, visn_attention_mask):
|
|
''' Cross Attention -- cross for vision not for language '''
|
|
visn_att_output, attention_scores = self.visual_attention(visn_input, lang_input, ctx_att_mask=lang_attention_mask)
|
|
return visn_att_output, attention_scores
|
|
|
|
def self_att(self, visn_input, visn_attention_mask):
|
|
''' Self Attention -- on visual features with language clues '''
|
|
visn_att_output = self.visn_self_att(visn_input, visn_attention_mask)
|
|
return visn_att_output
|
|
|
|
def output_fc(self, visn_input):
|
|
''' Feed forward '''
|
|
visn_inter_output = self.visn_inter(visn_input)
|
|
visn_output = self.visn_output(visn_inter_output, visn_input)
|
|
return visn_output
|
|
|
|
def forward(self, lang_feats, lang_attention_mask,
|
|
visn_feats, visn_attention_mask, tdx):
|
|
|
|
''' visual self-attention with state '''
|
|
visn_att_output = torch.cat((lang_feats[:, 0:1, :], visn_feats), dim=1)
|
|
state_vis_mask = torch.cat((lang_attention_mask[:,:,:,0:1], visn_attention_mask), dim=-1)
|
|
|
|
''' state and vision attend to language '''
|
|
visn_att_output, cross_attention_scores = self.cross_att(lang_feats[:, 1:, :], lang_attention_mask[:, :, :, 1:], visn_att_output, state_vis_mask)
|
|
|
|
language_attention_scores = cross_attention_scores[:, :, 0, :]
|
|
|
|
state_visn_att_output = self.self_att(visn_att_output, state_vis_mask)
|
|
state_visn_output = self.output_fc(state_visn_att_output[0])
|
|
|
|
visn_att_output = state_visn_output[:, 1:, :]
|
|
lang_att_output = torch.cat((state_visn_output[:, 0:1, :], lang_feats[:,1:,:]), dim=1)
|
|
|
|
visual_attention_scores = state_visn_att_output[1][:, :, 0, 1:]
|
|
|
|
return lang_att_output, visn_att_output, language_attention_scores, visual_attention_scores
|
|
|
|
|
|
class VisionEncoder(nn.Module):
|
|
def __init__(self, vision_size, config):
|
|
super().__init__()
|
|
feat_dim = vision_size
|
|
|
|
# Object feature encoding
|
|
self.visn_fc = nn.Linear(feat_dim, config.hidden_size)
|
|
self.visn_layer_norm = BertLayerNorm(config.hidden_size, eps=1e-12)
|
|
|
|
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
|
|
|
def forward(self, visn_input):
|
|
feats = visn_input
|
|
|
|
x = self.visn_fc(feats)
|
|
x = self.visn_layer_norm(x)
|
|
|
|
output = self.dropout(x)
|
|
return output
|
|
|
|
|
|
class VLNBert(BertPreTrainedModel):
|
|
def __init__(self, config):
|
|
super(VLNBert, self).__init__(config)
|
|
self.embeddings = BertEmbeddings(config)
|
|
self.pooler = BertPooler(config)
|
|
|
|
self.img_dim = config.img_feature_dim # 2176
|
|
logger.info('VLNBert Image Dimension: {}'.format(self.img_dim))
|
|
self.img_feature_type = config.img_feature_type # ''
|
|
self.vl_layers = config.vl_layers # 4
|
|
self.la_layers = config.la_layers # 9
|
|
self.lalayer = nn.ModuleList(
|
|
[BertLayer(config) for _ in range(self.la_layers)])
|
|
self.addlayer = nn.ModuleList(
|
|
[LXRTXLayer(config) for _ in range(self.vl_layers)])
|
|
self.vision_encoder = VisionEncoder(self.config.img_feature_dim, self.config)
|
|
self.apply(self.init_weights)
|
|
|
|
def forward(self, mode, input_ids, token_type_ids=None,
|
|
attention_mask=None, lang_mask=None, vis_mask=None, position_ids=None, head_mask=None, img_feats=None):
|
|
|
|
attention_mask = lang_mask
|
|
|
|
if token_type_ids is None:
|
|
token_type_ids = torch.zeros_like(input_ids)
|
|
|
|
extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
|
|
|
|
extended_attention_mask = extended_attention_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility
|
|
extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
|
|
|
|
head_mask = [None] * self.config.num_hidden_layers
|
|
|
|
if mode == 'language':
|
|
''' LXMERT language branch (in VLN only perform this at initialization) '''
|
|
embedding_output = self.embeddings(input_ids, position_ids=position_ids, token_type_ids=token_type_ids)
|
|
text_embeds = embedding_output
|
|
|
|
for layer_module in self.lalayer:
|
|
temp_output = layer_module(text_embeds, extended_attention_mask)
|
|
text_embeds = temp_output[0]
|
|
|
|
sequence_output = text_embeds
|
|
pooled_output = self.pooler(sequence_output)
|
|
|
|
return pooled_output, sequence_output
|
|
|
|
elif mode == 'visual':
|
|
''' LXMERT visual branch (no language processing during navigation) '''
|
|
text_embeds = input_ids
|
|
|
|
text_mask = extended_attention_mask
|
|
|
|
img_embedding_output = self.vision_encoder(img_feats)
|
|
img_seq_len = img_feats.shape[1]
|
|
batch_size = text_embeds.size(0)
|
|
|
|
img_seq_mask = vis_mask
|
|
|
|
extended_img_mask = img_seq_mask.unsqueeze(1).unsqueeze(2)
|
|
extended_img_mask = extended_img_mask.to(dtype=next(self.parameters()).dtype) # fp16 compatibility
|
|
extended_img_mask = (1.0 - extended_img_mask) * -10000.0
|
|
img_mask = extended_img_mask
|
|
|
|
lang_output = text_embeds
|
|
visn_output = img_embedding_output
|
|
|
|
for tdx, layer_module in enumerate(self.addlayer):
|
|
lang_output, visn_output, language_attention_scores, visual_attention_scores = layer_module(lang_output, text_mask, visn_output, img_mask, tdx)
|
|
|
|
sequence_output = lang_output
|
|
pooled_output = self.pooler(sequence_output)
|
|
|
|
language_state_scores = language_attention_scores.mean(dim=1)
|
|
visual_action_scores = visual_attention_scores.mean(dim=1)
|
|
|
|
# weighted_feat
|
|
language_attention_probs = nn.Softmax(dim=-1)(language_state_scores.clone()).unsqueeze(-1)
|
|
visual_attention_probs = nn.Softmax(dim=-1)(visual_action_scores.clone()).unsqueeze(-1)
|
|
|
|
attended_language = (language_attention_probs * text_embeds[:, 1:, :]).sum(1)
|
|
attended_visual = (visual_attention_probs * img_embedding_output).sum(1)
|
|
|
|
return pooled_output, visual_action_scores, attended_language, attended_visual
|