deviandice/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Spinenet: implementation without 4x downsampling right off the bat

Browse Source
This commit is contained in:
James Betker 2020-09-21 12:36:30 -06:00
parent 384e3d54cc
commit 9429544a60
1 changed files with 20 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
codes/models/archs/spinenet_arch.py
View File

@ -253,7 +253,8 @@ class SpineNet(nn.Module):
norm_cfg=dict(type='BN', requires_grad=True), norm_cfg=dict(type='BN', requires_grad=True),
zero_init_residual=True, zero_init_residual=True,
activation='relu', activation='relu',
use_input_norm=False): use_input_norm=False,
double_reduce_early=True):
super(SpineNet, self).__init__() super(SpineNet, self).__init__()
self._block_specs = build_block_specs()[2:] self._block_specs = build_block_specs()[2:]
self._endpoints_num_filters = SCALING_MAP[arch]['endpoints_num_filters'] self._endpoints_num_filters = SCALING_MAP[arch]['endpoints_num_filters']
@ -262,6 +263,7 @@ class SpineNet(nn.Module):
self._filter_size_scale = SCALING_MAP[arch]['filter_size_scale'] self._filter_size_scale = SCALING_MAP[arch]['filter_size_scale']
self._init_block_fn = Bottleneck self._init_block_fn = Bottleneck
self._num_init_blocks = 2 self._num_init_blocks = 2
self._early_double_reduce = double_reduce_early
self.zero_init_residual = zero_init_residual self.zero_init_residual = zero_init_residual
assert min(output_level) > 2 and max(output_level) < 8, "Output level out of range" assert min(output_level) > 2 and max(output_level) < 8, "Output level out of range"
self.output_level = output_level self.output_level = output_level
@ -274,12 +276,20 @@ class SpineNet(nn.Module):
def _make_stem_layer(self, in_channels): def _make_stem_layer(self, in_channels):
"""Build the stem network.""" """Build the stem network."""
# Build the first conv and maxpooling layers. # Build the first conv and maxpooling layers.
self.conv1 = ConvBnRelu( if self._early_double_reduce:
in_channels, self.conv1 = ConvBnRelu(
64, in_channels,
kernel_size=7, 64,
stride=2) kernel_size=7,
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) stride=2)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
else:
self.conv1 = ConvBnRelu(
in_channels,
64,
kernel_size=7,
stride=1)
self.maxpool = None
# Build the initial level 2 blocks. # Build the initial level 2 blocks.
self.init_block1 = make_res_layer( self.init_block1 = make_res_layer(
@ -340,7 +350,9 @@ class SpineNet(nn.Module):
std = torch.Tensor([0.229, 0.224, 0.225]).view(1, 3, 1, 1).to(input.device) std = torch.Tensor([0.229, 0.224, 0.225]).view(1, 3, 1, 1).to(input.device)
input = (input - mean) / std input = (input - mean) / std
feat = self.maxpool(self.conv1(input)) feat = self.conv1(input)
if self.maxpool:
feat = self.maxpool(feat)
feat1 = self.init_block1(feat) feat1 = self.init_block1(feat)
feat2 = self.init_block2(feat1) feat2 = self.init_block2(feat1)
block_feats = [feat1, feat2] block_feats = [feat1, feat2]