Inheritance diagram for fastreid.layers.frn.FRN:
Public Member Functions | |
| __init__ (self, num_features, eps=1e-6, is_eps_leanable=False) | |
| reset_parameters (self) | |
| extra_repr (self) | |
| forward (self, x) | |
Public Attributes | |
| num_features | |
| init_eps | |
| is_eps_leanable | |
| weight | |
| bias | |
| eps | |
Detailed Description
Constructor & Destructor Documentation
◆ __init__()
| fastreid.layers.frn.FRN.__init__ | ( | self, | |
| num_features, | |||
| eps = 1e-6, | |||
| is_eps_leanable = False ) |
weight = gamma, bias = beta
beta, gamma:
Variables of shape [1, 1, 1, C]. if TensorFlow
Variables of shape [1, C, 1, 1]. if PyTorch
eps: A scalar constant or learnable variable.
Definition at line 33 of file frn.py.
33 def __init__(self, num_features, eps=1e-6, is_eps_leanable=False):
34 """
35 weight = gamma, bias = beta
36 beta, gamma:
37 Variables of shape [1, 1, 1, C]. if TensorFlow
38 Variables of shape [1, C, 1, 1]. if PyTorch
39 eps: A scalar constant or learnable variable.
40 """
41 super(FRN, self).__init__()
42
43 self.num_features = num_features
44 self.init_eps = eps
45 self.is_eps_leanable = is_eps_leanable
46
47 self.weight = Parameter(torch.Tensor(num_features))
48 self.bias = Parameter(torch.Tensor(num_features))
49 if is_eps_leanable:
50 self.eps = Parameter(torch.Tensor(1))
51 else:
52 self.register_buffer('eps', torch.Tensor([eps]))
53 self.reset_parameters()
54
Member Function Documentation
◆ extra_repr()
| fastreid.layers.frn.FRN.extra_repr | ( | self | ) |
◆ forward()
| fastreid.layers.frn.FRN.forward | ( | self, | |
| x ) |
0, 1, 2, 3 -> (B, H, W, C) in TensorFlow
0, 1, 2, 3 -> (B, C, H, W) in PyTorch
TensorFlow code
nu2 = tf.reduce_mean(tf.square(x), axis=[1, 2], keepdims=True)
x = x * tf.rsqrt(nu2 + tf.abs(eps))
# This Code include TLU function max(y, tau)
return tf.maximum(gamma * x + beta, tau)
Definition at line 64 of file frn.py.
64 def forward(self, x):
65 """
66 0, 1, 2, 3 -> (B, H, W, C) in TensorFlow
67 0, 1, 2, 3 -> (B, C, H, W) in PyTorch
68 TensorFlow code
69 nu2 = tf.reduce_mean(tf.square(x), axis=[1, 2], keepdims=True)
70 x = x * tf.rsqrt(nu2 + tf.abs(eps))
71 # This Code include TLU function max(y, tau)
72 return tf.maximum(gamma * x + beta, tau)
73 """
74 # Compute the mean norm of activations per channel.
75 nu2 = x.pow(2).mean(dim=[2, 3], keepdim=True)
76
77 # Perform FRN.
78 x = x * torch.rsqrt(nu2 + self.eps.abs())
79
80 # Scale and Bias
81 x = self.weight.view(1, self.num_features, 1, 1) * x + self.bias.view(1, self.num_features, 1, 1)
82 # x = self.weight * x + self.bias
83 return x
84
85
◆ reset_parameters()
| fastreid.layers.frn.FRN.reset_parameters | ( | self | ) |
Member Data Documentation
◆ bias
◆ eps
◆ init_eps
◆ is_eps_leanable
◆ num_features
◆ weight
The documentation for this class was generated from the following file:
- smreid/fastreid/layers/frn.py
