目录
- 模型构建
- FFN
End-to-End Object Detection with Transformers(DETR)
论文地址:https://arxiv.org/abs/2005.12872
源代码位置: https://github.com/facebookresearch/detr
参考文献: javascript:void(0)
模型整体结构
论文中模型结构
主干网络
- backbone(CNN-Resnet)
- CNN网络
- positional(位置信息)
- transformer
- encoder
- decoder
- predicttion head
模型构建
models/detr.py
# 构建两大模型
# backbone = build_backbone(args)
# transformer = build_transformer(args)
# 模型连接 DETR
#
def build(args):
num_classes = 20 if args.dataset_file != 'coco' else 91
if args.dataset_file == "coco_panoptic":
# for panoptic, we just add a num_classes that is large enough to hold
# max_obj_id + 1, but the exact value doesn't really matter
num_classes = 250
device = torch.device(args.device)
# 包含两大部分, 构建 backbone 和 构建 transformer
backbone = build_backbone(args)
transformer = build_transformer(args)
model = DETR(
backbone,
transformer,
num_classes=num_classes,
num_queries=args.num_queries,
aux_loss=args.aux_loss,
)
if args.masks:
model = DETRsegm(model, freeze_detr=(args.frozen_weights is not None))
backbone
cnn骨架特征提取
backbone的输入和输出
- input shape=(N,3,W,H)
- output shape=(N,2048,W/32,H/32) #针对 Resnet50 C=2048, 针对 Resnet18,Resnet34 C=512
假设输入是(N,C,H,W),则resnet50输出是(N,2048,H//32,W//32),1024比较大,
为了节省计算量,先采用1x1卷积降维为256,(hidden_dim=256,在main.py 中设置参数)
最后转化为序列格式输入到transformer中,输入shape=(H*W,N,256),H=H/32,W=W/32
class Backbone(BackboneBase):
"""ResNet backbone with frozen BatchNorm."""
def __init__(self, name: str,
train_backbone: bool,
return_interm_layers: bool,
dilation: bool):
backbone = getattr(torchvision.models, name)(
replace_stride_with_dilation=[False, False, dilation],
pretrained=is_main_process(), norm_layer=FrozenBatchNorm2d)
# 针对不同的网络,选择了不同的输出大小
num_channels = 512 if name in ('resnet18', 'resnet34') else 2048
super().__init__(backbone, train_backbone, num_channels, return_interm_layers)
···
```python
在 DETR 类中
src 为 backone 的输出 shape=(N,512,W/32,H/32)
# self.input_proj(src) 将 shape=(N,512,W/32,H/32) -> shape=(N,256,W/32,H/32)
hs = self.transformer(self.input_proj(src), mask, self.query_embed.weight, pos[-1])[0]
位置信息标注,包含了x,y两个方向的位置信息。编码方式任然采用sincos, 语音序列只是包含了一个方向的位置信息
PositionEmbeddingSine.forward的输入和输出
- input NestedTensor型数据 tensor_list的类型是NestedTensor,内部自动附加了mask,
- x.tensors.shape=((N, 512,W/32, H/32) x.mask.shape=(N,W/32,H/32)
- output: pos.shape=(N, 256, W/32,H/32)
class PositionEmbeddingSine(nn.Module):
"""
This is a more standard version of the position embedding, very similar to the one
used by the Attention is all you need paper, generalized to work on images.
"""
def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None):
super().__init__()
self.num_pos_feats = num_pos_feats
self.temperature = temperature
self.normalize = normalize
if scale is not None and normalize is False:
raise ValueError("normalize should be True if scale is passed")
if scale is None:
scale = 2 * math.pi
self.scale = scale
def forward(self, tensor_list: NestedTensor):
x = tensor_list.tensors
mask = tensor_list.mask
#x.tensors.shape=((N, 512,W/32, H/32) x.mask.shape=(N,W/32,H/32)
assert mask is not None
not_mask = ~mask
y_embed = not_mask.cumsum(1, dtype=torch.float32)
x_embed = not_mask.cumsum(2, dtype=torch.float32)
if self.normalize:
eps = 1e-6
y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
# 前面输入向量是256,编码是一半sin,一半cos
dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
dim_t = self.temperature ** (2 * (dim_t // 2) / self.num_pos_feats)
pos_x = x_embed[:, :, :, None] / dim_t
pos_y = y_embed[:, :, :, None] / dim_t
pos_x = torch.stack((pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4).flatten(3)
pos_y = torch.stack((pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4).flatten(3)
pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2)
# pos.shape=(N, 256, W/32,H/32) 前128是y方向编码,而128是x方向编码
return pos
transformer
transformer整体构建
model/transformer.py
Transformer 模型构建
包含 encoder decoder
class Transformer(nn.Module):
def __init__(self, d_model=512, nhead=8, num_encoder_layers=6,
num_decoder_layers=6, dim_feedforward=2048, dropout=0.1,
activation="relu", normalize_before=False,
return_intermediate_dec=False):
super().__init__()
# 编码
encoder_layer = TransformerEncoderLayer(d_model, nhead, dim_feedforward,
dropout, activation, normalize_before)
encoder_norm = nn.LayerNorm(d_model) if normalize_before else None
self.encoder = TransformerEncoder(encoder_layer, num_encoder_layers, encoder_norm)
# 解码
decoder_layer = TransformerDecoderLayer(d_model, nhead, dim_feedforward,
dropout, activation, normalize_before)
decoder_norm = nn.LayerNorm(d_model)
self.decoder = TransformerDecoder(decoder_layer, num_decoder_layers, decoder_norm,
return_intermediate=return_intermediate_dec)
self._reset_parameters()
self.d_model = d_model
self.nhead = nhead
def _reset_parameters(self):
for p in self.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
def forward(self, src, mask, query_embed, pos_embed):
# flatten NxCxHxW to HWxNxC
# inputs: {src,mask,query_embed,pos} 由 DETR.forward 获取来自 backbone
bs, c, h, w = src.shape
# 先对数据做变换
# 特殊说明 这里是经过backbone 输出的特征 (N,256,W/32,H/32) 之后transformer过程中 输出shape为(H/32xW/32,N,256) 特征的宽和高没有变化,为了书写方法方便,我这里将 W/32,H/32 写成为 W,H
# src=(N,256,W/32,H/32)-> (WH,N,256)
# pos_embed=(N,256,W,H)-> (WH,N,256)
# query_embed=(100,256) -> (100,N,256)
# mask=(N,W,H) -> (N,WH)
src = src.flatten(2).permute(2, 0, 1)
pos_embed = pos_embed.flatten(2).permute(2, 0, 1)
query_embed = query_embed.unsqueeze(1).repeat(1, bs, 1)
mask = mask.flatten(1)
# 解码 第一层 首次参数设置为0,后续自动更新
tgt = torch.zeros_like(query_embed)
# encoder src=(WH,N,256) mask= (N,WH) pos_embed= (WH,N,256)
# 输出 (WH,N,256)
memory = self.encoder(src, src_key_padding_mask=mask, pos=pos_embed)
# decoder tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH)
# pos_embed=(WH,N,256) query_embed=(100,N,256)
# 输出 hs=(decoder_layers, 100, N, 256)
hs = self.decoder(tgt, memory, memory_key_padding_mask=mask,
pos=pos_embed, query_pos=query_embed)
# return (decoder_layers, N, 100, 256) (N, 256, H, W])
return hs.transpose(1, 2), memory.permute(1, 2, 0).view(bs, c, h, w)
Encoder
编码器结构和输入输出
编码器的输入有三个 src=(WH,N,256) src_mask= (N,WH) pos_embed= (WH,N,256) 注释:W=W/32,H=H/32
- 由图像生成的序列,shape=(WH,N,256)
- 掩码信息,shape= (N,WH)
- 图像序列的空间位置信息,shape=(WH,N,256)
经过6层编码后 输出只有一个 序列,shape和输入的src 序列保持一直,shape=(WH,N,256) 注释:W=W/32,H=H/32
模型细节
- 原始transformer的n个编码器输入中,只有第一个编码器需要输入位置编码向量,但是DETR里面对每个编码器都输入了同一个位置编码向量
- QKV处理逻辑不同,在编码器内部位置编码仅仅和 Q K 相加,V 不做任何处理
TransformerEncoder类
def _get_clones(module, N):
return nn.ModuleList([copy.deepcopy(module) for i in range(N)])
class TransformerEncoder(nn.Module):
def __init__(self, encoder_layer, num_layers, norm=None):
super().__init__()
self.layers = _get_clones(encoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm
def forward(self, src,
mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None):
output = src
# 默认设置了 6个 编码器,循环6遍
# encoder input src=(WH,N,256) src_mask= (N,WH) pos_embed= (WH,N,256)
# output -> output (WH,N,256)
# 包含了多层相同的结构,首尾相连,上一层输出为下一层的输入
for layer in self.layers:
output = layer(output, src_mask=mask,
src_key_padding_mask=src_key_padding_mask, pos=pos)
if self.norm is not None:
output = self.norm(output)
return output
TransformerEncoderLayer类
class TransformerEncoderLayer(nn.Module):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1,
activation="relu", normalize_before=False):
super().__init__()
self.self_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout)
# Implementation of Feedforward model
self.linear1 = nn.Linear(d_model, dim_feedforward)
self.dropout = nn.Dropout(dropout)
self.linear2 = nn.Linear(dim_feedforward, d_model)
self.norm1 = nn.LayerNorm(d_model)
self.norm2 = nn.LayerNorm(d_model)
self.dropout1 = nn.Dropout(dropout)
self.dropout2 = nn.Dropout(dropout)
self.activation = _get_activation_fn(activation)
self.normalize_before = normalize_before
def with_pos_embed(self, tensor, pos: Optional[Tensor]):
return tensor if pos is None else tensor + pos
def forward_post(self,
src,
src_mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None):
# src=(WH,N,256) mask= (N,WH) pos_embed= (WH,N,256)
# with_pos_embed 输入是 src pos {图片序列,位置信息}
# 对 Q K 进行更新
q = k = self.with_pos_embed(src, pos)
# MultiheadAttention 多头注意力机制
# 在编码器内部位置编码仅仅和QK相加,V不做任何处理
src2 = self.self_attn(q, k, value=src, attn_mask=src_mask,
key_padding_mask=src_key_padding_mask)[0]
# 残差
src = src + self.dropout1(src2)
src = self.norm1(src)
# FFN
src2 = self.linear2(self.dropout(self.activation(self.linear1(src))))
src = src + self.dropout2(src2)
src = self.norm2(src)
return src
def forward_pre(self, src,
src_mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None):
src2 = self.norm1(src)
q = k = self.with_pos_embed(src2, pos)
src2 = self.self_attn(q, k, value=src2, attn_mask=src_mask,
key_padding_mask=src_key_padding_mask)[0]
src = src + self.dropout1(src2)
src2 = self.norm2(src)
src2 = self.linear2(self.dropout(self.activation(self.linear1(src2))))
src = src + self.dropout2(src2)
return src
def forward(self, src,
src_mask: Optional[Tensor] = None,
src_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None):
# encoder src=(WH,N,256) mask= (N,WH) pos_embed= (WH,N,256)
# output=(WH,N,256)
# 默认 normalize_before=False 只对 forward_post 函数注解
if self.normalize_before:
return self.forward_pre(src, src_mask, src_key_padding_mask, pos)
return self.forward_post(src, src_mask, src_key_padding_mask, pos)
Decoder
解码器结构和输入输出
输入参数
解码器的输入 有五个参数 decoder tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH) pos_embed=(WH,N,256) query_pos=(100,N,256)
- tgt 可以理解为上一层解码器的解码输出 shape=(100,N,256) 第一层的tgt=torch.zeros_like(query_embed) 为零矩阵
- memory 最后一个编码器输出 shape=(WH,N,256)
- mask 掩码信息 shape=(N,WH)
- pos 和编码器输入中完全相同位置参数 shape=(WH,N,256)
- query_pos 是可学习输出位置向量, 个人理解 解码器中的这个参数 全局共享 提供全局注意力 query_pos=(100,N,256)
输出参数
- 输出 (decoder_layers, 100, N, 256) decoder_layers 为解码器的数量(层数),原文默认设置为6层
原始transformer顺序解码操作不同的是,detr一次就把N个无序框并行输出
Obeject Query
针对 query_pos 参数的其他博客解释
论文中指出object queries作用非常类似faster rcnn中的anchor,只不过这里是可学习的,不是提前设置好的。
object queries(shape是(100,256)) 源代码中,这是一个torch.nn.Embedding的对象。
官方介绍:一个保存了固定字典和大小的简单查找表。这个模块常用来保存词嵌入和用下标检索它们。模块的输入是一个下标的列表,输出是对应的词嵌入。
个人理解:query_pos 可以简单认为是输出位置编码,其作用主要是在学习过程中提供目标对象和全局图像之间的关系,相当于全局注意力,必不可少非常关键。代码形式上是可学习位置编码矩阵。和编码器一样,该可学习位置编码向量也会输入到每一个解码器中。我们可以尝试通俗理解:object queries矩阵内部通过学习建模了100个物体之间的全局关系,并且参与到网络的学习当中。
其他细节:
- tgt(第一次输入是query embeding,第二次是上一层的输出out);
- 和编码器一样,只是Q 与 K加上了位置编码信息, V不会加入位置编码
- 引入可学习的Object queries
- 不需要顺序解码,一次即可输出N个无序集合
TransformerDecoder类
class TransformerDecoder(nn.Module):
def __init__(self, decoder_layer, num_layers, norm=None, return_intermediate=False):
super().__init__()
self.layers = _get_clones(decoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm
self.return_intermediate = return_intermediate
def forward(self, tgt, memory,
tgt_mask: Optional[Tensor] = None,
memory_mask: Optional[Tensor] = None,
tgt_key_padding_mask: Optional[Tensor] = None,
memory_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None,
query_pos: Optional[Tensor] = None):
# decoder tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH) pos_embed=(WH,N,256) query_embed=(100,N,256)
output = tgt
intermediate = []
for layer in self.layers:
output = layer(output, memory, tgt_mask=tgt_mask,
memory_mask=memory_mask,
tgt_key_padding_mask=tgt_key_padding_mask,
memory_key_padding_mask=memory_key_padding_mask,
pos=pos, query_pos=query_pos)
if self.return_intermediate:
intermediate.append(self.norm(output))
if self.norm is not None:
output = self.norm(output)
if self.return_intermediate:
intermediate.pop()
intermediate.append(output)
# intermediate=[outpout...] intermediate[0].shape=(100,N,256)
# return_intermediate = True
if self.return_intermediate:
return torch.stack(intermediate)
return output.unsqueeze(0)
TransformerDecoderLayer类
class TransformerDecoderLayer(nn.Module):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1,
activation="relu", normalize_before=False):
super().__init__()
self.self_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.multihead_attn = nn.MultiheadAttention(d_model, nhead, dropout=dropout)
# Implementation of Feedforward model
self.linear1 = nn.Linear(d_model, dim_feedforward)
self.dropout = nn.Dropout(dropout)
self.linear2 = nn.Linear(dim_feedforward, d_model)
self.norm1 = nn.LayerNorm(d_model)
self.norm2 = nn.LayerNorm(d_model)
self.norm3 = nn.LayerNorm(d_model)
self.dropout1 = nn.Dropout(dropout)
self.dropout2 = nn.Dropout(dropout)
self.dropout3 = nn.Dropout(dropout)
self.activation = _get_activation_fn(activation)
self.normalize_before = normalize_before
def with_pos_embed(self, tensor, pos: Optional[Tensor]):
return tensor if pos is None else tensor + pos
def forward_post(self, tgt, memory,
tgt_mask: Optional[Tensor] = None,
memory_mask: Optional[Tensor] = None,
tgt_key_padding_mask: Optional[Tensor] = None,
memory_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None,
query_pos: Optional[Tensor] = None):
# # decoder tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH) pos_embed=(WH,N,256) query_embed=(100,N,256)
# 解码 第一次注意力机制 tgt=(100,N,256) 是 上一个单元输出 如果是第一次 torch.zeros_like(query_embed)
# query_embed=(100,N,256) query_pos 应该是共享单元,不管多少层都是公用一组数据
q = k = self.with_pos_embed(tgt, query_pos)
tgt2 = self.self_attn(q, k, value=tgt, attn_mask=tgt_mask,
key_padding_mask=tgt_key_padding_mask)[0]
tgt = tgt + self.dropout1(tgt2)
tgt = self.norm1(tgt)
# multihead_attn
# query=self.with_pos_embed(tgt, query_pos) 在第二次注意力机制中 对 Q 进行更新
# key=self.with_pos_embed(memory, pos) 在第二次注意力机制中对 K 进行更新
tgt2 = self.multihead_attn(query=self.with_pos_embed(tgt, query_pos),
key=self.with_pos_embed(memory, pos),
value=memory, attn_mask=memory_mask,
key_padding_mask=memory_key_padding_mask)[0]
tgt = tgt + self.dropout2(tgt2)
tgt = self.norm2(tgt)
# FFN
tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt))))
tgt = tgt + self.dropout3(tgt2)
tgt = self.norm3(tgt)
return tgt
def forward_pre(self, tgt, memory,
tgt_mask: Optional[Tensor] = None,
memory_mask: Optional[Tensor] = None,
tgt_key_padding_mask: Optional[Tensor] = None,
memory_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None,
query_pos: Optional[Tensor] = None):
# # decoder tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH) pos_embed=(WH,N,256) query_embed=(100,N,256)
tgt2 = self.norm1(tgt)
q = k = self.with_pos_embed(tgt2, query_pos)
tgt2 = self.self_attn(q, k, value=tgt2, attn_mask=tgt_mask,
key_padding_mask=tgt_key_padding_mask)[0]
tgt = tgt + self.dropout1(tgt2)
tgt2 = self.norm2(tgt)
tgt2 = self.multihead_attn(query=self.with_pos_embed(tgt2, query_pos),
key=self.with_pos_embed(memory, pos),
value=memory, attn_mask=memory_mask,
key_padding_mask=memory_key_padding_mask)[0]
tgt = tgt + self.dropout2(tgt2)
tgt2 = self.norm3(tgt)
tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt2))))
tgt = tgt + self.dropout3(tgt2)
return tgt
def forward(self, tgt, memory,
tgt_mask: Optional[Tensor] = None,
memory_mask: Optional[Tensor] = None,
tgt_key_padding_mask: Optional[Tensor] = None,
memory_key_padding_mask: Optional[Tensor] = None,
pos: Optional[Tensor] = None,
query_pos: Optional[Tensor] = None):
# decoder input tgt=(100,N,256) memory=(WH,N,256),mask=(N,WH) pos_embed=(WH,N,256) query_embed=(100,N,256)
# ISFalse
if self.normalize_before:
return self.forward_pre(tgt, memory, tgt_mask, memory_mask,
tgt_key_padding_mask, memory_key_padding_mask, pos, query_pos)
return self.forward_post(tgt, memory, tgt_mask, memory_mask,
tgt_key_padding_mask, memory_key_padding_mask, pos, query_pos)
FFN
最后是接了一个FFN,就是两个全连接层,一个用于分类,一个用于回归预测
分类: 一层模型结构
最终预测 MLP模型 是由具有ReLU激活功能且具有隐藏层的3层感知器和线性层计算的。 FFN预测框的标准化中心坐标,高度和宽度, 输入图像,然后线性层使用softmax函数预测类标签
DETR类中
# 输入 hs.shape = (decoder_layers, N, 100, 256)
# 分类 self.class_embed = nn.Linear(hidden_dim, num_classes + 1)
# FFN Linear class input=(decoder_layers, N, 100, 256) output=(decoder_layers, N, 100, num_classes+1)
outputs_class = self.class_embed(hs)
# 预测 self.bbox_embed = MLP(hidden_dim, hidden_dim, 4, 3)
# MLP Bounding box input=(decoder_layers, N, 100, 256) output=(decoder_layers, N, 100, 4)
outputs_coord = self.bbox_embed(hs).sigmoid()