Spaces:
Runtime error
Runtime error
File size: 4,713 Bytes
3e99b05 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 |
# coding=utf-8
# Copyright 2022 The IDEA Authors. 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.
import torch
import torch.nn as nn
from .attention import MultiheadAttention
class OriginalConditionalAttentionEncoder(nn.Module):
"""Original implementation of Conditional Self-Attention
Remove norm and dropout layer for test simplicity
"""
def __init__(self, d_model, nhead):
super().__init__()
self.sa_qcontent_proj = nn.Linear(d_model, d_model)
self.sa_qpos_proj = nn.Linear(d_model, d_model)
self.sa_kcontent_proj = nn.Linear(d_model, d_model)
self.sa_kpos_proj = nn.Linear(d_model, d_model)
self.sa_v_proj = nn.Linear(d_model, d_model)
self.self_attn = MultiheadAttention(d_model, nhead, dropout=0.0, vdim=d_model)
def forward(self, tgt, query_pos):
q_content = self.sa_qcontent_proj(tgt)
q_pos = self.sa_qpos_proj(query_pos)
k_content = self.sa_kcontent_proj(tgt)
k_pos = self.sa_kpos_proj(query_pos)
v = self.sa_v_proj(tgt)
q = q_content + q_pos
k = k_content + k_pos
tgt2 = self.self_attn(q, k, v)
return tgt2
class OriginalConditionalAttentionDecoder(nn.Module):
"""Original implementation of Conditional Attention Decoder
Remove norm and dropout layer for test simplicity
"""
def __init__(self, d_model, nhead):
super().__init__()
# self-attn
self.sa_qcontent_proj = nn.Linear(d_model, d_model)
self.sa_qpos_proj = nn.Linear(d_model, d_model)
self.sa_kcontent_proj = nn.Linear(d_model, d_model)
self.sa_kpos_proj = nn.Linear(d_model, d_model)
self.sa_v_proj = nn.Linear(d_model, d_model)
self.self_attn = MultiheadAttention(d_model, nhead, dropout=0.0, vdim=d_model)
# cross-attn
self.ca_qcontent_proj = nn.Linear(d_model, d_model)
self.ca_qpos_proj = nn.Linear(d_model, d_model)
self.ca_kcontent_proj = nn.Linear(d_model, d_model)
self.ca_kpos_proj = nn.Linear(d_model, d_model)
self.ca_v_proj = nn.Linear(d_model, d_model)
self.ca_qpos_sine_proj = nn.Linear(d_model, d_model)
self.cross_attn = MultiheadAttention(d_model * 2, nhead, dropout=0.0, vdim=d_model)
self.nhead = nhead
def forward(self, tgt, memory, query_pos, pos, query_sine_embed, is_first=True):
# self attention
q_content = self.sa_qcontent_proj(
tgt
) # target is the input of the first decoder layer. zero by default.
q_pos = self.sa_qpos_proj(query_pos)
k_content = self.sa_kcontent_proj(tgt)
k_pos = self.sa_kpos_proj(query_pos)
v = self.sa_v_proj(tgt)
q = q_content + q_pos
k = k_content + k_pos
tgt2 = self.self_attn(q, k, v)[0]
tgt = tgt + tgt2
# ========================================
# cross attention
q_content = self.ca_qcontent_proj(tgt)
k_content = self.ca_kcontent_proj(memory)
v = self.ca_v_proj(memory)
num_queries, bs, n_model = q_content.shape
hw, _, _ = k_content.shape
k_pos = self.ca_kpos_proj(pos)
# For the first decoder layer, we concatenate the positional embedding predicted from
# the object query (the positional embedding) into the original query (key) in DETR.
if is_first:
q_pos = self.ca_qpos_proj(query_pos)
q = q_content + q_pos
k = k_content + k_pos
else:
q = q_content
k = k_content
q = q.view(num_queries, bs, self.nhead, n_model // self.nhead)
query_sine_embed = self.ca_qpos_sine_proj(query_sine_embed)
query_sine_embed = query_sine_embed.view(num_queries, bs, self.nhead, n_model // self.nhead)
q = torch.cat([q, query_sine_embed], dim=3).view(num_queries, bs, n_model * 2)
k = k.view(hw, bs, self.nhead, n_model // self.nhead)
k_pos = k_pos.view(hw, bs, self.nhead, n_model // self.nhead)
k = torch.cat([k, k_pos], dim=3).view(hw, bs, n_model * 2)
tgt2 = self.cross_attn(query=q, key=k, value=v)[0]
return tgt2 + tgt
|