Spaces:
Running
on
Zero
Running
on
Zero
File size: 3,627 Bytes
e85fecb |
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 |
"""
Copied from RT-DETR (https://github.com/lyuwenyu/RT-DETR)
Copyright(c) 2023 lyuwenyu. All Rights Reserved.
"""
import torch
import torch.nn as nn
class ConvNormLayer(nn.Module):
def __init__(self, ch_in, ch_out, kernel_size, stride, padding=None, bias=False, act=None):
super().__init__()
self.conv = nn.Conv2d(
ch_in,
ch_out,
kernel_size,
stride,
padding=(kernel_size - 1) // 2 if padding is None else padding,
bias=bias,
)
self.norm = nn.BatchNorm2d(ch_out)
self.act = nn.Identity() if act is None else get_activation(act)
def forward(self, x):
return self.act(self.norm(self.conv(x)))
class FrozenBatchNorm2d(nn.Module):
"""copy and modified from https://github.com/facebookresearch/detr/blob/master/models/backbone.py
BatchNorm2d where the batch statistics and the affine parameters are fixed.
Copy-paste from torchvision.misc.ops with added eps before rqsrt,
without which any other models than torchvision.models.resnet[18,34,50,101]
produce nans.
"""
def __init__(self, num_features, eps=1e-5):
super(FrozenBatchNorm2d, self).__init__()
n = num_features
self.register_buffer("weight", torch.ones(n))
self.register_buffer("bias", torch.zeros(n))
self.register_buffer("running_mean", torch.zeros(n))
self.register_buffer("running_var", torch.ones(n))
self.eps = eps
self.num_features = n
def _load_from_state_dict(
self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
):
num_batches_tracked_key = prefix + "num_batches_tracked"
if num_batches_tracked_key in state_dict:
del state_dict[num_batches_tracked_key]
super(FrozenBatchNorm2d, self)._load_from_state_dict(
state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
)
def forward(self, x):
# move reshapes to the beginning
# to make it fuser-friendly
w = self.weight.reshape(1, -1, 1, 1)
b = self.bias.reshape(1, -1, 1, 1)
rv = self.running_var.reshape(1, -1, 1, 1)
rm = self.running_mean.reshape(1, -1, 1, 1)
scale = w * (rv + self.eps).rsqrt()
bias = b - rm * scale
return x * scale + bias
def extra_repr(self):
return "{num_features}, eps={eps}".format(**self.__dict__)
def freeze_batch_norm2d(module: nn.Module) -> nn.Module:
if isinstance(module, nn.BatchNorm2d):
module = FrozenBatchNorm2d(module.num_features)
else:
for name, child in module.named_children():
_child = freeze_batch_norm2d(child)
if _child is not child:
setattr(module, name, _child)
return module
def get_activation(act: str, inplace: bool = True):
"""get activation"""
if act is None:
return nn.Identity()
elif isinstance(act, nn.Module):
return act
act = act.lower()
if act == "silu" or act == "swish":
m = nn.SiLU()
elif act == "relu":
m = nn.ReLU()
elif act == "leaky_relu":
m = nn.LeakyReLU()
elif act == "silu":
m = nn.SiLU()
elif act == "gelu":
m = nn.GELU()
elif act == "hardsigmoid":
m = nn.Hardsigmoid()
else:
raise RuntimeError("")
if hasattr(m, "inplace"):
m.inplace = inplace
return m
|