|
|
import copy |
|
|
import torch |
|
|
import torch.nn as nn |
|
|
from ...builder import TRANSFORMERS |
|
|
from ..layers import BaseTransformerLayer, MultiheadAttention, FFN |
|
|
|
|
|
|
|
|
@TRANSFORMERS.register_module() |
|
|
class DETRDecoder(nn.Module): |
|
|
def __init__( |
|
|
self, |
|
|
embed_dim=256, |
|
|
num_heads=8, |
|
|
attn_dropout=0.1, |
|
|
ffn_dim=2048, |
|
|
ffn_dropout=0.1, |
|
|
num_layers=6, |
|
|
post_norm=True, |
|
|
batch_first=True, |
|
|
return_intermediate=True, |
|
|
): |
|
|
super().__init__() |
|
|
|
|
|
self.embed_dim = embed_dim |
|
|
self.num_heads = num_heads |
|
|
self.attn_dropout = attn_dropout |
|
|
self.ffn_dim = ffn_dim |
|
|
self.ffn_dropout = ffn_dropout |
|
|
self.num_layers = num_layers |
|
|
self.batch_first = batch_first |
|
|
self.return_intermediate = return_intermediate |
|
|
|
|
|
self._init_transformer_layers() |
|
|
|
|
|
if post_norm: |
|
|
self.post_norm_layer = nn.LayerNorm(self.embed_dim) |
|
|
else: |
|
|
self.post_norm_layer = None |
|
|
|
|
|
def _init_transformer_layers(self): |
|
|
transformer_layers = BaseTransformerLayer( |
|
|
attn=MultiheadAttention( |
|
|
embed_dim=self.embed_dim, |
|
|
num_heads=self.num_heads, |
|
|
attn_drop=self.attn_dropout, |
|
|
batch_first=self.batch_first, |
|
|
), |
|
|
ffn=FFN( |
|
|
embed_dim=self.embed_dim, |
|
|
ffn_dim=self.ffn_dim, |
|
|
ffn_drop=self.ffn_dropout, |
|
|
), |
|
|
norm=nn.LayerNorm( |
|
|
normalized_shape=self.embed_dim, |
|
|
), |
|
|
operation_order=("self_attn", "norm", "cross_attn", "norm", "ffn", "norm"), |
|
|
) |
|
|
self.layers = nn.ModuleList([copy.deepcopy(transformer_layers) for _ in range(self.num_layers)]) |
|
|
|
|
|
def forward( |
|
|
self, |
|
|
query, |
|
|
key, |
|
|
value, |
|
|
query_pos=None, |
|
|
key_pos=None, |
|
|
attn_masks=None, |
|
|
query_key_padding_mask=None, |
|
|
key_padding_mask=None, |
|
|
**kwargs, |
|
|
): |
|
|
if not self.return_intermediate: |
|
|
for layer in self.layers: |
|
|
query = layer( |
|
|
query, |
|
|
key, |
|
|
value, |
|
|
query_pos=query_pos, |
|
|
key_pos=key_pos, |
|
|
attn_masks=attn_masks, |
|
|
query_key_padding_mask=query_key_padding_mask, |
|
|
key_padding_mask=key_padding_mask, |
|
|
**kwargs, |
|
|
) |
|
|
|
|
|
if self.post_norm_layer is not None: |
|
|
query = self.post_norm_layer(query)[None] |
|
|
return query |
|
|
|
|
|
|
|
|
intermediate = [] |
|
|
for layer in self.layers: |
|
|
query = layer( |
|
|
query, |
|
|
key, |
|
|
value, |
|
|
query_pos=query_pos, |
|
|
key_pos=key_pos, |
|
|
attn_masks=attn_masks, |
|
|
query_key_padding_mask=query_key_padding_mask, |
|
|
key_padding_mask=key_padding_mask, |
|
|
**kwargs, |
|
|
) |
|
|
|
|
|
if self.return_intermediate: |
|
|
if self.post_norm_layer is not None: |
|
|
intermediate.append(self.post_norm_layer(query)) |
|
|
else: |
|
|
intermediate.append(query) |
|
|
|
|
|
return torch.stack(intermediate) |
|
|
|
