File size: 9,815 Bytes
3eb682b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 |
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
"""ResNe(X)t 3D stem helper."""
import torch.nn as nn
def get_stem_func(name):
"""
Retrieves the stem module by name.
"""
trans_funcs = {"x3d_stem": X3DStem, "basic_stem": ResNetBasicStem}
assert (
name in trans_funcs.keys()
), "Transformation function '{}' not supported".format(name)
return trans_funcs[name]
class VideoModelStem(nn.Module):
"""
Video 3D stem module. Provides stem operations of Conv, BN, ReLU, MaxPool
on input data tensor for one or multiple pathways.
"""
def __init__(
self,
dim_in,
dim_out,
kernel,
stride,
padding,
inplace_relu=True,
eps=1e-5,
bn_mmt=0.1,
norm_module=nn.BatchNorm3d,
stem_func_name="basic_stem",
):
"""
The `__init__` method of any subclass should also contain these
arguments. List size of 1 for single pathway models (C2D, I3D, Slow
and etc), list size of 2 for two pathway models (SlowFast).
Args:
dim_in (list): the list of channel dimensions of the inputs.
dim_out (list): the output dimension of the convolution in the stem
layer.
kernel (list): the kernels' size of the convolutions in the stem
layers. Temporal kernel size, height kernel size, width kernel
size in order.
stride (list): the stride sizes of the convolutions in the stem
layer. Temporal kernel stride, height kernel size, width kernel
size in order.
padding (list): the paddings' sizes of the convolutions in the stem
layer. Temporal padding size, height padding size, width padding
size in order.
inplace_relu (bool): calculate the relu on the original input
without allocating new memory.
eps (float): epsilon for batch norm.
bn_mmt (float): momentum for batch norm. Noted that BN momentum in
PyTorch = 1 - BN momentum in Caffe2.
norm_module (nn.Module): nn.Module for the normalization layer. The
default is nn.BatchNorm3d.
stem_func_name (string): name of the the stem function applied on
input to the network.
"""
super(VideoModelStem, self).__init__()
assert (
len(
{
len(dim_in),
len(dim_out),
len(kernel),
len(stride),
len(padding),
}
)
== 1
), "Input pathway dimensions are not consistent."
self.num_pathways = len(dim_in)
self.kernel = kernel
self.stride = stride
self.padding = padding
self.inplace_relu = inplace_relu
self.eps = eps
self.bn_mmt = bn_mmt
# Construct the stem layer.
self._construct_stem(dim_in, dim_out, norm_module, stem_func_name)
def _construct_stem(self, dim_in, dim_out, norm_module, stem_func_name):
trans_func = get_stem_func(stem_func_name)
for pathway in range(len(dim_in)):
stem = trans_func(
dim_in[pathway],
dim_out[pathway],
self.kernel[pathway],
self.stride[pathway],
self.padding[pathway],
self.inplace_relu,
self.eps,
self.bn_mmt,
norm_module,
)
self.add_module("pathway{}_stem".format(pathway), stem)
def forward(self, x):
assert (
len(x) == self.num_pathways
), "Input tensor does not contain {} pathway".format(self.num_pathways)
for pathway in range(len(x)):
m = getattr(self, "pathway{}_stem".format(pathway))
x[pathway] = m(x[pathway])
return x
class ResNetBasicStem(nn.Module):
"""
ResNe(X)t 3D stem module.
Performs spatiotemporal Convolution, BN, and Relu following by a
spatiotemporal pooling.
"""
def __init__(
self,
dim_in,
dim_out,
kernel,
stride,
padding,
inplace_relu=True,
eps=1e-5,
bn_mmt=0.1,
norm_module=nn.BatchNorm3d,
):
"""
The `__init__` method of any subclass should also contain these arguments.
Args:
dim_in (int): the channel dimension of the input. Normally 3 is used
for rgb input, and 2 or 3 is used for optical flow input.
dim_out (int): the output dimension of the convolution in the stem
layer.
kernel (list): the kernel size of the convolution in the stem layer.
temporal kernel size, height kernel size, width kernel size in
order.
stride (list): the stride size of the convolution in the stem layer.
temporal kernel stride, height kernel size, width kernel size in
order.
padding (int): the padding size of the convolution in the stem
layer, temporal padding size, height padding size, width
padding size in order.
inplace_relu (bool): calculate the relu on the original input
without allocating new memory.
eps (float): epsilon for batch norm.
bn_mmt (float): momentum for batch norm. Noted that BN momentum in
PyTorch = 1 - BN momentum in Caffe2.
norm_module (nn.Module): nn.Module for the normalization layer. The
default is nn.BatchNorm3d.
"""
super(ResNetBasicStem, self).__init__()
self.kernel = kernel
self.stride = stride
self.padding = padding
self.inplace_relu = inplace_relu
self.eps = eps
self.bn_mmt = bn_mmt
# Construct the stem layer.
self._construct_stem(dim_in, dim_out, norm_module)
def _construct_stem(self, dim_in, dim_out, norm_module):
self.conv = nn.Conv3d(
dim_in,
dim_out,
self.kernel,
stride=self.stride,
padding=self.padding,
bias=False,
)
self.bn = norm_module(
num_features=dim_out, eps=self.eps, momentum=self.bn_mmt
)
self.relu = nn.ReLU(self.inplace_relu)
self.pool_layer = nn.MaxPool3d(
kernel_size=[1, 3, 3], stride=[1, 2, 2], padding=[0, 1, 1]
)
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
x = self.relu(x)
x = self.pool_layer(x)
return x
class X3DStem(nn.Module):
"""
X3D's 3D stem module.
Performs a spatial followed by a depthwise temporal Convolution, BN, and Relu following by a
spatiotemporal pooling.
"""
def __init__(
self,
dim_in,
dim_out,
kernel,
stride,
padding,
inplace_relu=True,
eps=1e-5,
bn_mmt=0.1,
norm_module=nn.BatchNorm3d,
):
"""
The `__init__` method of any subclass should also contain these arguments.
Args:
dim_in (int): the channel dimension of the input. Normally 3 is used
for rgb input, and 2 or 3 is used for optical flow input.
dim_out (int): the output dimension of the convolution in the stem
layer.
kernel (list): the kernel size of the convolution in the stem layer.
temporal kernel size, height kernel size, width kernel size in
order.
stride (list): the stride size of the convolution in the stem layer.
temporal kernel stride, height kernel size, width kernel size in
order.
padding (int): the padding size of the convolution in the stem
layer, temporal padding size, height padding size, width
padding size in order.
inplace_relu (bool): calculate the relu on the original input
without allocating new memory.
eps (float): epsilon for batch norm.
bn_mmt (float): momentum for batch norm. Noted that BN momentum in
PyTorch = 1 - BN momentum in Caffe2.
norm_module (nn.Module): nn.Module for the normalization layer. The
default is nn.BatchNorm3d.
"""
super(X3DStem, self).__init__()
self.kernel = kernel
self.stride = stride
self.padding = padding
self.inplace_relu = inplace_relu
self.eps = eps
self.bn_mmt = bn_mmt
# Construct the stem layer.
self._construct_stem(dim_in, dim_out, norm_module)
def _construct_stem(self, dim_in, dim_out, norm_module):
self.conv_xy = nn.Conv3d(
dim_in,
dim_out,
kernel_size=(1, self.kernel[1], self.kernel[2]),
stride=(1, self.stride[1], self.stride[2]),
padding=(0, self.padding[1], self.padding[2]),
bias=False,
)
self.conv = nn.Conv3d(
dim_out,
dim_out,
kernel_size=(self.kernel[0], 1, 1),
stride=(self.stride[0], 1, 1),
padding=(self.padding[0], 0, 0),
bias=False,
groups=dim_out,
)
self.bn = norm_module(
num_features=dim_out, eps=self.eps, momentum=self.bn_mmt
)
self.relu = nn.ReLU(self.inplace_relu)
def forward(self, x):
x = self.conv_xy(x)
x = self.conv(x)
x = self.bn(x)
x = self.relu(x)
return x
|