Inheritance diagram for i3d_fusion.I3DFusion:
Public Member Functions | |
| __init__ (self, in_channels, out_channels, dropout_ratio, input_key) | |
| forward (self, feat_dict) | |
Public Attributes | |
| out_channels | |
| dropout_ratio | |
| input_key | |
| avg_pool | |
| fc | |
| bn | |
| relu | |
| dropout | |
Detailed Description
3차원 입력 텐서들을 연결하는 기능의 모듈
입력 텐서들을 풀링하여 하나의 텐서로 이어붙이고 1차원으로 변환함, 추가적으로 fc+bn+relu 연산을 적용할 수 있음
args:
in_channels (int) : 입력 텐서 채널들의 합
out_channels (int) : 출력 텐서의 채널 수, 0 이상값을 설정하면 fc+bn+relu 연산을 적용함
dropout_ratio (float) : fc+bn+relu 연산을 사용할 경우 드랍아웃 비율을 설정하는 파라미터
input_key (str) : inference에 사용하는 입력데이터의 키값들
Definition at line 4 of file i3d_fusion.py.
Constructor & Destructor Documentation
◆ __init__()
| i3d_fusion.I3DFusion.__init__ | ( | self, | |
| in_channels, | |||
| out_channels, | |||
| dropout_ratio, | |||
| input_key ) |
Definition at line 14 of file i3d_fusion.py.
14 def __init__(self, in_channels, out_channels, dropout_ratio, input_key):
15 super().__init__()
16 self.out_channels = out_channels
17 self.dropout_ratio = dropout_ratio
18 self.input_key = input_key
19 self.avg_pool = nn.AdaptiveAvgPool3d((1, 1, 1))
20
21 if self.out_channels > 0:
22 self.fc = nn.Linear(in_channels, out_channels)
23 self.bn = nn.BatchNorm1d(out_channels)
24 self.relu = nn.ReLU(inplace=True)
25
26 if self.dropout_ratio != 0:
27 self.dropout = nn.Dropout(p=self.dropout_ratio)
28 else:
29 self.dropout = None
30
Member Function Documentation
◆ forward()
| i3d_fusion.I3DFusion.forward | ( | self, | |
| feat_dict ) |
args:
sample (dict)) : 입력 데이터, self.input_key에 해당하는 키가 있어야함
self.input_key의 아이템은 Tensor 타입 -> shape (B, C, T, H, W)
B : 배치 크기
C : 입력 채널
T : 시간
return (Tensor):
1차원 특징 벡터 -> shape (B, C_o)
B : 배치 크기
C_o : 채널 크기, out_channels 또는 입력 채널의 합
Definition at line 31 of file i3d_fusion.py.
31 def forward(self, feat_dict):
32 """
33 args:
34 sample (dict)) : 입력 데이터, self.input_key에 해당하는 키가 있어야함
35 self.input_key의 아이템은 Tensor 타입 -> shape (B, C, T, H, W)
36 B : 배치 크기
37 C : 입력 채널
38 T : 시간
39
40 return (Tensor):
41 1차원 특징 벡터 -> shape (B, C_o)
42 B : 배치 크기
43 C_o : 채널 크기, out_channels 또는 입력 채널의 합
44 """
45 feats = []
46 for key in self.input_key:
47 x = feat_dict[key]
48 x = self.avg_pool(x)
49 feats.append(x)
50
51 fusion_feat = torch.cat(feats, dim=1)
52 fusion_feat = fusion_feat.view(x.shape[0], -1)
53
54 if self.out_channels > 0:
55 if self.dropout is not None:
56 fusion_feat = self.dropout(fusion_feat)
57
58 fusion_feat = self.relu(self.bn(self.fc(fusion_feat)))
59
60 return fusion_feat
61 else:
62 return fusion_feat
63
Member Data Documentation
◆ avg_pool
| i3d_fusion.I3DFusion.avg_pool |
Definition at line 19 of file i3d_fusion.py.
◆ bn
| i3d_fusion.I3DFusion.bn |
Definition at line 23 of file i3d_fusion.py.
◆ dropout
| i3d_fusion.I3DFusion.dropout |
Definition at line 27 of file i3d_fusion.py.
◆ dropout_ratio
| i3d_fusion.I3DFusion.dropout_ratio |
Definition at line 17 of file i3d_fusion.py.
◆ fc
| i3d_fusion.I3DFusion.fc |
Definition at line 22 of file i3d_fusion.py.
◆ input_key
| i3d_fusion.I3DFusion.input_key |
Definition at line 18 of file i3d_fusion.py.
◆ out_channels
| i3d_fusion.I3DFusion.out_channels |
Definition at line 16 of file i3d_fusion.py.
◆ relu
| i3d_fusion.I3DFusion.relu |
Definition at line 24 of file i3d_fusion.py.
The documentation for this class was generated from the following file:
- smaction/models/heads/i3d_fusion.py
