Merge branch 'select_kernel' into attention

timm/models/__init__.py

@@ -16,6 +16,7 @@ from .gluon_xception import *
 from .res2net import *
 from .dla import *
 from .hrnet import *
+from .sknet import *
 
 from .registry import *
 from .factory import create_model

timm/models/conv2d_layers.py

@@ -1,3 +1,5 @@
+from collections import OrderedDict
+
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -100,14 +102,11 @@ def create_conv2d_pad(in_chs, out_chs, kernel_size, **kwargs):
         return nn.Conv2d(in_chs, out_chs, kernel_size, padding=padding, **kwargs)
 
 
-class MixedConv2d(nn.Module):
+class MixedConv2d(nn.ModuleDict):
     """ Mixed Grouped Convolution
     Based on MDConv and GroupedConv in MixNet impl:
       https://github.com/tensorflow/tpu/blob/master/models/official/mnasnet/mixnet/custom_layers.py
-
-    NOTE: This does not currently work with torch.jit.script
     """
-
     def __init__(self, in_channels, out_channels, kernel_size=3,
                  stride=1, padding='', dilation=1, depthwise=False, **kwargs):
         super(MixedConv2d, self).__init__()
@@ -131,7 +130,7 @@ class MixedConv2d(nn.Module):
 
     def forward(self, x):
         x_split = torch.split(x, self.splits, 1)
-        x_out = [c(x) for x, c in zip(x_split, self._modules.values())]
+        x_out = [c(x_split[i]) for i, c in enumerate(self.values())]
         x = torch.cat(x_out, 1)
         return x
 
@@ -240,6 +239,110 @@ class CondConv2d(nn.Module):
         return out
 
 
+class SelectiveKernelAttn(nn.Module):
+    def __init__(self, channels, num_paths=2, attn_channels=32,
+                 act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+        super(SelectiveKernelAttn, self).__init__()
+        self.num_paths = num_paths
+        self.pool = nn.AdaptiveAvgPool2d(1)
+        self.fc_reduce = nn.Conv2d(channels, attn_channels, kernel_size=1, bias=False)
+        self.bn = norm_layer(attn_channels)
+        self.act = act_layer(inplace=True)
+        self.fc_select = nn.Conv2d(attn_channels, channels * num_paths, kernel_size=1, bias=False)
+
+    def forward(self, x):
+        assert x.shape[1] == self.num_paths
+        x = torch.sum(x, dim=1)
+        x = self.pool(x)
+        x = self.fc_reduce(x)
+        x = self.bn(x)
+        x = self.act(x)
+        x = self.fc_select(x)
+        B, C, H, W = x.shape
+        x = x.view(B, self.num_paths, C // self.num_paths, H, W)
+        x = torch.softmax(x, dim=1)
+        return x
+
+
+def _kernel_valid(k):
+    if isinstance(k, (list, tuple)):
+        for ki in k:
+            return _kernel_valid(ki)
+    assert k >= 3 and k % 2
+
+
+class ConvBnAct(nn.Module):
+    def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, dilation=1, groups=1,
+                 drop_block=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+        super(ConvBnAct, self).__init__()
+        padding = _get_padding(kernel_size, stride, dilation)  # assuming PyTorch style padding for this block
+        self.conv = nn.Conv2d(
+            in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride,
+            padding=padding, dilation=dilation, groups=groups, bias=False)
+        self.bn = norm_layer(out_channels)
+        self.drop_block = drop_block
+        if act_layer is not None:
+            self.act = act_layer(inplace=True)
+        else:
+            self.act = None
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
+        if self.act is not None:
+            x = self.act(x)
+        return x
+
+
+class SelectiveKernelConv(nn.Module):
+
+    def __init__(self, in_channels, out_channels, kernel_size=None, stride=1, dilation=1, groups=1,
+                 attn_reduction=16, min_attn_channels=32, keep_3x3=True, split_input=False,
+                 drop_block=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+        super(SelectiveKernelConv, self).__init__()
+        kernel_size = kernel_size or [3, 5]
+        _kernel_valid(kernel_size)
+        if not isinstance(kernel_size, list):
+            kernel_size = [kernel_size] * 2
+        if keep_3x3:
+            dilation = [dilation * (k - 1) // 2 for k in kernel_size]
+            kernel_size = [3] * len(kernel_size)
+        else:
+            dilation = [dilation] * len(kernel_size)
+        self.num_paths = len(kernel_size)
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.split_input = split_input
+        if self.split_input:
+            assert in_channels % self.num_paths == 0
+            in_channels = in_channels // self.num_paths
+        groups = min(out_channels, groups)
+
+        conv_kwargs = dict(
+            stride=stride, groups=groups, drop_block=drop_block, act_layer=act_layer, norm_layer=norm_layer)
+        self.paths = nn.ModuleList([
+            ConvBnAct(in_channels, out_channels, kernel_size=k, dilation=d, **conv_kwargs)
+            for k, d in zip(kernel_size, dilation)])
+
+        attn_channels = max(int(out_channels / attn_reduction), min_attn_channels)
+        self.attn = SelectiveKernelAttn(out_channels, self.num_paths, attn_channels)
+        self.drop_block = drop_block
+
+    def forward(self, x):
+        if self.split_input:
+            x_split = torch.split(x, self.in_channels // self.num_paths, 1)
+            x_paths = [op(x_split[i]) for i, op in enumerate(self.paths)]
+        else:
+            x_paths = [op(x) for op in self.paths]
+        x = torch.stack(x_paths, dim=1)
+        x_attn = self.attn(x)
+        x = x * x_attn
+        x = torch.sum(x, dim=1)
+        return x
+
+
 # helper method
 def select_conv2d(in_chs, out_chs, kernel_size, **kwargs):
     assert 'groups' not in kwargs  # only use 'depthwise' bool arg
@@ -256,5 +359,3 @@ def select_conv2d(in_chs, out_chs, kernel_size, **kwargs):
         else:
             m = create_conv2d_pad(in_chs, out_chs, kernel_size, groups=groups, **kwargs)
     return m
-
-

timm/models/nn_ops.py

@@ -0,0 +1,77 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+import math
+
+
+def drop_block_2d(x, drop_prob=0.1, block_size=7, gamma_scale=1.0, drop_with_noise=False):
+    _, _, height, width = x.shape
+    total_size = width * height
+    clipped_block_size = min(block_size, min(width, height))
+    # seed_drop_rate, the gamma parameter
+    seed_drop_rate = gamma_scale * drop_prob * total_size / clipped_block_size ** 2 / (
+            (width - block_size + 1) *
+            (height - block_size + 1))
+
+    # Forces the block to be inside the feature map.
+    w_i, h_i = torch.meshgrid(torch.arange(width).to(x.device), torch.arange(height).to(x.device))
+    valid_block = ((w_i >= clipped_block_size // 2) & (w_i < width - (clipped_block_size - 1) // 2)) & \
+                  ((h_i >= clipped_block_size // 2) & (h_i < height - (clipped_block_size - 1) // 2))
+    valid_block = torch.reshape(valid_block, (1, 1, height, width)).float()
+
+    uniform_noise = torch.rand_like(x, dtype=torch.float32)
+    block_mask = ((2 - seed_drop_rate - valid_block + uniform_noise) >= 1).float()
+    block_mask = -F.max_pool2d(
+        -block_mask,
+        kernel_size=clipped_block_size,  # block_size,
+        stride=1,
+        padding=clipped_block_size // 2)
+
+    if drop_with_noise:
+        normal_noise = torch.randn_like(x)
+        x = x * block_mask + normal_noise * (1 - block_mask)
+    else:
+        normalize_scale = block_mask.numel() / (torch.sum(block_mask) + 1e-7)
+        x = x * block_mask * normalize_scale
+    return x
+
+
+class DropBlock2d(nn.Module):
+    """ DropBlock. See https://arxiv.org/pdf/1810.12890.pdf
+    """
+    def __init__(self,
+                 drop_prob=0.1,
+                 block_size=7,
+                 gamma_scale=1.0,
+                 with_noise=False):
+        super(DropBlock2d, self).__init__()
+        self.drop_prob = drop_prob
+        self.gamma_scale = gamma_scale
+        self.block_size = block_size
+        self.with_noise = with_noise
+
+    def forward(self, x):
+        if not self.training or not self.drop_prob:
+            return x
+        return drop_block_2d(x, self.drop_prob, self.block_size, self.gamma_scale, self.with_noise)
+
+
+def drop_path(x, drop_prob=0.):
+    """Drop paths (Stochastic Depth) per sample (when applied in residual blocks)."""
+    keep_prob = 1 - drop_prob
+    random_tensor = keep_prob + torch.rand((x.size()[0], 1, 1, 1), dtype=x.dtype, device=x.device)
+    random_tensor.floor_()  # binarize
+    output = x.div(keep_prob) * random_tensor
+    return output
+
+
+class DropPath(nn.ModuleDict):
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        if not self.training or not self.drop_prob:
+            return x
+        return drop_path(x, self.drop_prob)

timm/models/res2net.py

@@ -54,14 +54,15 @@ class Bottle2neck(nn.Module):
 
     def __init__(self, inplanes, planes, stride=1, downsample=None,
                  cardinality=1, base_width=26, scale=4, use_se=False,
-                 act_layer=nn.ReLU, norm_layer=None, dilation=1, previous_dilation=1, **_):
+                 act_layer=nn.ReLU, norm_layer=None, dilation=1, first_dilation=None, **_):
         super(Bottle2neck, self).__init__()
         self.scale = scale
         self.is_first = stride > 1 or downsample is not None
         self.num_scales = max(1, scale - 1)
         width = int(math.floor(planes * (base_width / 64.0))) * cardinality
-        outplanes = planes * self.expansion
         self.width = width
+        outplanes = planes * self.expansion
+        first_dilation = first_dilation or dilation
 
         self.conv1 = nn.Conv2d(inplanes, width * scale, kernel_size=1, bias=False)
         self.bn1 = norm_layer(width * scale)
@@ -70,8 +71,8 @@ class Bottle2neck(nn.Module):
         bns = []
         for i in range(self.num_scales):
             convs.append(nn.Conv2d(
-                width, width, kernel_size=3, stride=stride, padding=dilation,
-                dilation=dilation, groups=cardinality, bias=False))
+                width, width, kernel_size=3, stride=stride, padding=first_dilation,
+                dilation=first_dilation, groups=cardinality, bias=False))
             bns.append(norm_layer(width))
         self.convs = nn.ModuleList(convs)
         self.bns = nn.ModuleList(bns)
@@ -86,6 +87,9 @@ class Bottle2neck(nn.Module):
         self.relu = act_layer(inplace=True)
         self.downsample = downsample
 
+    def zero_init_last_bn(self):
+        nn.init.zeros_(self.bn3.weight)
+
     def forward(self, x):
         residual = x
 

timm/models/resnet.py

@@ -15,6 +15,7 @@ from .registry import register_model
 from .helpers import load_pretrained
 from .adaptive_avgmax_pool import SelectAdaptivePool2d
 from .layers import EcaModule
+from .nn_ops import DropBlock2d, DropPath
 from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 
 
@@ -108,7 +109,7 @@ default_cfgs = {
 }
 
 
-def _get_padding(kernel_size, stride, dilation=1):
+def get_padding(kernel_size, stride, dilation=1):
     padding = ((stride - 1) + dilation * (kernel_size - 1)) // 2
     return padding
 
@@ -136,115 +137,135 @@ class BasicBlock(nn.Module):
     __constants__ = ['se', 'downsample']  # for pre 1.4 torchscript compat
     expansion = 1
 
-    def __init__(self, inplanes, planes, stride=1, downsample=None,
-                 cardinality=1, base_width=64, use_se=False, use_eca = False,
-                 reduce_first=1, dilation=1, previous_dilation=1, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+    def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64, use_se=False,
+                 reduce_first=1, dilation=1, first_dilation=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d,
+                 drop_block=None, drop_path=None):
         super(BasicBlock, self).__init__()
 
         assert cardinality == 1, 'BasicBlock only supports cardinality of 1'
         assert base_width == 64, 'BasicBlock doest not support changing base width'
         first_planes = planes // reduce_first
         outplanes = planes * self.expansion
+        first_dilation = first_dilation or dilation
 
         self.conv1 = nn.Conv2d(
-            inplanes, first_planes, kernel_size=3, stride=stride, padding=dilation,
-            dilation=dilation, bias=False)
+            inplanes, first_planes, kernel_size=3, stride=stride, padding=first_dilation,
+            dilation=first_dilation, bias=False)
         self.bn1 = norm_layer(first_planes)
         self.act1 = act_layer(inplace=True)
         self.conv2 = nn.Conv2d(
-            first_planes, outplanes, kernel_size=3, padding=previous_dilation,
-            dilation=previous_dilation, bias=False)
+            first_planes, outplanes, kernel_size=3, padding=dilation, dilation=dilation, bias=False)
         self.bn2 = norm_layer(outplanes)
 
         self.se = SEModule(outplanes, planes // 4) if use_se else None
-        self.eca = EcaModule(outplanes) if use_eca else None
 
         self.act2 = act_layer(inplace=True)
         self.downsample = downsample
         self.stride = stride
         self.dilation = dilation
+        self.drop_block = drop_block
+        self.drop_path = drop_path
+
+    def zero_init_last_bn(self):
+        nn.init.zeros_(self.bn2.weight)
 
     def forward(self, x):
         residual = x
 
-        out = self.conv1(x)
-        out = self.bn1(out)
-        out = self.act1(out)
-        out = self.conv2(out)
-        out = self.bn2(out)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
+        x = self.act1(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
 
         if self.se is not None:
-            out = self.se(out)
-        if self.eca is not None:
-            out = self.eca(out)
+            x = self.se(x)
 
-        if self.downsample is not None:
-            residual = self.downsample(x)
+        if self.drop_path is not None:
+            x = self.drop_path(x)
 
-        out += residual
-        out = self.act2(out)
+        if self.downsample is not None:
+            residual = self.downsample(residual)
+        x += residual
+        x = self.act2(x)
 
-        return out
+        return x
 
 
 class Bottleneck(nn.Module):
     __constants__ = ['se', 'downsample']  # for pre 1.4 torchscript compat
     expansion = 4
 
-    def __init__(self, inplanes, planes, stride=1, downsample=None,
-                 cardinality=1, base_width=64, use_se=False, use_eca=False,
-                 reduce_first=1, dilation=1, previous_dilation=1, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+    def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64, use_se=False,
+                 reduce_first=1, dilation=1, first_dilation=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d,
+                 drop_block=None, drop_path=None):
         super(Bottleneck, self).__init__()
 
         width = int(math.floor(planes * (base_width / 64)) * cardinality)
         first_planes = width // reduce_first
         outplanes = planes * self.expansion
+        first_dilation = first_dilation or dilation
 
         self.conv1 = nn.Conv2d(inplanes, first_planes, kernel_size=1, bias=False)
         self.bn1 = norm_layer(first_planes)
         self.act1 = act_layer(inplace=True)
         self.conv2 = nn.Conv2d(
             first_planes, width, kernel_size=3, stride=stride,
-            padding=dilation, dilation=dilation, groups=cardinality, bias=False)
+            padding=first_dilation, dilation=first_dilation, groups=cardinality, bias=False)
         self.bn2 = norm_layer(width)
         self.act2 = act_layer(inplace=True)
         self.conv3 = nn.Conv2d(width, outplanes, kernel_size=1, bias=False)
         self.bn3 = norm_layer(outplanes)
 
         self.se = SEModule(outplanes, planes // 4) if use_se else None
-        self.eca = EcaModule(outplanes) if use_eca else None
-    
+
         self.act3 = act_layer(inplace=True)
         self.downsample = downsample
         self.stride = stride
         self.dilation = dilation
+        self.drop_block = drop_block
+        self.drop_path = drop_path
+
+    def zero_init_last_bn(self):
+        nn.init.zeros_(self.bn3.weight)
 
     def forward(self, x):
         residual = x
 
-        out = self.conv1(x)
-        out = self.bn1(out)
-        out = self.act1(out)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
+        x = self.act1(x)
 
-        out = self.conv2(out)
-        out = self.bn2(out)
-        out = self.act2(out)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
+        x = self.act2(x)
 
-        out = self.conv3(out)
-        out = self.bn3(out)
+        x = self.conv3(x)
+        x = self.bn3(x)
+        if self.drop_block is not None:
+            x = self.drop_block(x)
 
         if self.se is not None:
-            out = self.se(out)
-        if self.eca is not None:
-            out = self.eca(out)
+            x = self.se(x)
 
-        if self.downsample is not None:
-            residual = self.downsample(x)
+        if self.drop_path is not None:
+            x = self.drop_path(x)
 
-        out += residual
-        out = self.act3(out)
+        if self.downsample is not None:
+            residual = self.downsample(residual)
+        x += residual
+        x = self.act3(x)
 
-        return out
+        return x
 
 
 class ResNet(nn.Module):
@@ -290,8 +311,6 @@ class ResNet(nn.Module):
         Number of input (color) channels.
     use_se : bool, default False
         Enable Squeeze-Excitation module in blocks
-    use_eca : bool, default False
-        Enable ECA module in blocks
     cardinality : int, default 1
         Number of convolution groups for 3x3 conv in Bottleneck.
     base_width : int, default 64
@@ -323,8 +342,8 @@ class ResNet(nn.Module):
     def __init__(self, block, layers, num_classes=1000, in_chans=3, use_se=False, use_eca=False,
                  cardinality=1, base_width=64, stem_width=64, stem_type='',
                  block_reduce_first=1, down_kernel_size=1, avg_down=False, output_stride=32,
-                 act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d, drop_rate=0.0, global_pool='avg',
-                 zero_init_last_bn=True, block_args=None):
+                 act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d, drop_rate=0.0, drop_path_rate=0.,
+                 drop_block_rate=0., global_pool='avg', zero_init_last_bn=True, block_args=None):
         block_args = block_args or dict()
         self.num_classes = num_classes
         deep_stem = 'deep' in stem_type
@@ -356,6 +375,9 @@ class ResNet(nn.Module):
         self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
 
         # Feature Blocks
+        dp = DropPath(drop_path_rate) if drop_block_rate else None
+        db_3 = DropBlock2d(drop_block_rate, 7, 0.25) if drop_block_rate else None
+        db_4 = DropBlock2d(drop_block_rate, 7, 1.00) if drop_block_rate else None
         channels, strides, dilations = [64, 128, 256, 512], [1, 2, 2, 2], [1] * 4
         if output_stride == 16:
             strides[3] = 1
@@ -367,29 +389,28 @@ class ResNet(nn.Module):
             assert output_stride == 32
         llargs = list(zip(channels, layers, strides, dilations))
         lkwargs = dict(
-            use_se=use_se, use_eca=use_eca, reduce_first=block_reduce_first, act_layer=act_layer, norm_layer=norm_layer,
-            avg_down=avg_down, down_kernel_size=down_kernel_size, **block_args)
+            use_se=use_se, reduce_first=block_reduce_first, act_layer=act_layer, norm_layer=norm_layer,
+            avg_down=avg_down, down_kernel_size=down_kernel_size, drop_path=dp, **block_args)
         self.layer1 = self._make_layer(block, *llargs[0], **lkwargs)
         self.layer2 = self._make_layer(block, *llargs[1], **lkwargs)
-        self.layer3 = self._make_layer(block, *llargs[2], **lkwargs)
-        self.layer4 = self._make_layer(block, *llargs[3], **lkwargs)
+        self.layer3 = self._make_layer(block, drop_block=db_3, *llargs[2], **lkwargs)
+        self.layer4 = self._make_layer(block, drop_block=db_4, *llargs[3], **lkwargs)
 
         # Head (Pooling and Classifier)
         self.global_pool = SelectAdaptivePool2d(pool_type=global_pool)
         self.num_features = 512 * block.expansion
         self.fc = nn.Linear(self.num_features * self.global_pool.feat_mult(), num_classes)
 
-        last_bn_name = 'bn3' if 'Bottle' in block.__name__ else 'bn2'
         for n, m in self.named_modules():
             if isinstance(m, nn.Conv2d):
                 nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
             elif isinstance(m, nn.BatchNorm2d):
-                if zero_init_last_bn and 'layer' in n and last_bn_name in n:
-                    # Initialize weight/gamma of last BN in each residual block to zero
-                    nn.init.constant_(m.weight, 0.)
-                else:
-                    nn.init.constant_(m.weight, 1.)
+                nn.init.constant_(m.weight, 1.)
                 nn.init.constant_(m.bias, 0.)
+        if zero_init_last_bn:
+            for m in self.modules():
+                if hasattr(m, 'zero_init_last_bn'):
+                    m.zero_init_last_bn()
 
     def _make_layer(self, block, planes, blocks, stride=1, dilation=1, reduce_first=1,
                     use_se=False, use_eca=False,avg_down=False, down_kernel_size=1, **kwargs):
@@ -397,7 +418,7 @@ class ResNet(nn.Module):
         downsample = None
         down_kernel_size = 1 if stride == 1 and dilation == 1 else down_kernel_size
         if stride != 1 or self.inplanes != planes * block.expansion:
-            downsample_padding = _get_padding(down_kernel_size, stride)
+            downsample_padding = get_padding(down_kernel_size, stride)
             downsample_layers = []
             conv_stride = stride
             if avg_down:
@@ -413,13 +434,10 @@ class ResNet(nn.Module):
         first_dilation = 1 if dilation in (1, 2) else 2
         bkwargs = dict(
             cardinality=self.cardinality, base_width=self.base_width, reduce_first=reduce_first,
-            use_se=use_se, use_eca=use_eca, **kwargs)
-        layers = [block(
-            self.inplanes, planes, stride, downsample, dilation=first_dilation, previous_dilation=dilation, **bkwargs)]
+            dilation=dilation, use_se=use_se, **kwargs)
+        layers = [block(self.inplanes, planes, stride, downsample, first_dilation=first_dilation, **bkwargs)]
         self.inplanes = planes * block.expansion
-        for i in range(1, blocks):
-            layers.append(block(
-                self.inplanes, planes, dilation=dilation, previous_dilation=dilation, **bkwargs))
+        layers += [block(self.inplanes, planes, **bkwargs) for _ in range(1, blocks)]
 
         return nn.Sequential(*layers)
 
@@ -447,8 +465,8 @@ class ResNet(nn.Module):
     def forward(self, x):
         x = self.forward_features(x)
         x = self.global_pool(x).flatten(1)
-        if self.drop_rate > 0.:
-            x = F.dropout(x, p=self.drop_rate, training=self.training)
+        if self.drop_rate:
+            x = F.dropout(x, p=float(self.drop_rate), training=self.training)
         x = self.fc(x)
         return x
 

timm/models/sknet.py

@@ -0,0 +1,242 @@
+import math
+
+from torch import nn as nn
+
+from timm.models.registry import register_model
+from timm.models.helpers import load_pretrained
+from timm.models.conv2d_layers import SelectiveKernelConv, ConvBnAct
+from timm.models.resnet import ResNet, SEModule
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+
+
+def _cfg(url='', **kwargs):
+    return {
+        'url': url,
+        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
+        'crop_pct': 0.875, 'interpolation': 'bilinear',
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
+        'first_conv': 'conv1', 'classifier': 'fc',
+        **kwargs
+    }
+
+
+default_cfgs = {
+    'skresnet18': _cfg(url=''),
+    'skresnet26d': _cfg(),
+    'skresnet50': _cfg(),
+    'skresnet50d': _cfg(),
+    'skresnext50_32x4d': _cfg(),
+}
+
+
+class SelectiveKernelBasic(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64,
+                 use_se=False, sk_kwargs=None, reduce_first=1, dilation=1, first_dilation=None,
+                 drop_block=None, drop_path=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+        super(SelectiveKernelBasic, self).__init__()
+
+        sk_kwargs = sk_kwargs or {}
+        conv_kwargs = dict(drop_block=drop_block, act_layer=act_layer, norm_layer=norm_layer)
+        assert cardinality == 1, 'BasicBlock only supports cardinality of 1'
+        assert base_width == 64, 'BasicBlock doest not support changing base width'
+        first_planes = planes // reduce_first
+        out_planes = planes * self.expansion
+        first_dilation = first_dilation or dilation
+
+        _selective_first = True  # FIXME temporary, for experiments
+        if _selective_first:
+            self.conv1 = SelectiveKernelConv(
+                inplanes, first_planes, stride=stride, dilation=first_dilation, **conv_kwargs, **sk_kwargs)
+            conv_kwargs['act_layer'] = None
+            self.conv2 = ConvBnAct(
+                first_planes, out_planes, kernel_size=3, dilation=dilation, **conv_kwargs)
+        else:
+            self.conv1 = ConvBnAct(
+                inplanes, first_planes, kernel_size=3, stride=stride, dilation=first_dilation, **conv_kwargs)
+            conv_kwargs['act_layer'] = None
+            self.conv2 = SelectiveKernelConv(
+                first_planes, out_planes, dilation=dilation, **conv_kwargs, **sk_kwargs)
+        self.se = SEModule(out_planes, planes // 4) if use_se else None
+        self.act = act_layer(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+        self.dilation = dilation
+        self.drop_block = drop_block
+        self.drop_path = drop_path
+
+    def zero_init_last_bn(self):
+        nn.init.zeros_(self.conv2.bn.weight)
+
+    def forward(self, x):
+        residual = x
+        x = self.conv1(x)
+        x = self.conv2(x)
+        if self.se is not None:
+            x = self.se(x)
+        if self.drop_path is not None:
+            x = self.drop_path(x)
+        if self.downsample is not None:
+            residual = self.downsample(residual)
+        x += residual
+        x = self.act(x)
+        return x
+
+
+class SelectiveKernelBottleneck(nn.Module):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None,
+                 cardinality=1, base_width=64, use_se=False, sk_kwargs=None,
+                 reduce_first=1, dilation=1, first_dilation=None,
+                 drop_block=None, drop_path=None,
+                 act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
+        super(SelectiveKernelBottleneck, self).__init__()
+
+        sk_kwargs = sk_kwargs or {}
+        conv_kwargs = dict(drop_block=drop_block, act_layer=act_layer, norm_layer=norm_layer)
+        width = int(math.floor(planes * (base_width / 64)) * cardinality)
+        first_planes = width // reduce_first
+        out_planes = planes * self.expansion
+        first_dilation = first_dilation or dilation
+
+        self.conv1 = ConvBnAct(inplanes, first_planes, kernel_size=1, **conv_kwargs)
+        self.conv2 = SelectiveKernelConv(
+            first_planes, width, stride=stride, dilation=first_dilation, groups=cardinality,
+            **conv_kwargs, **sk_kwargs)
+        conv_kwargs['act_layer'] = None
+        self.conv3 = ConvBnAct(width, out_planes, kernel_size=1, **conv_kwargs)
+        self.se = SEModule(out_planes, planes // 4) if use_se else None
+        self.act = act_layer(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+        self.dilation = dilation
+        self.drop_block = drop_block
+        self.drop_path = drop_path
+
+    def zero_init_last_bn(self):
+        nn.init.zeros_(self.conv3.bn.weight)
+
+    def forward(self, x):
+        residual = x
+        x = self.conv1(x)
+        x = self.conv2(x)
+        x = self.conv3(x)
+        if self.se is not None:
+            x = self.se(x)
+        if self.drop_path is not None:
+            x = self.drop_path(x)
+        if self.downsample is not None:
+            residual = self.downsample(residual)
+        x += residual
+        x = self.act(x)
+        return x
+
+
+@register_model
+def skresnet18(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a ResNet-18 model.
+    """
+    default_cfg = default_cfgs['skresnet18']
+    sk_kwargs = dict(
+        min_attn_channels=16,
+    )
+    model = ResNet(
+        SelectiveKernelBasic, [2, 2, 2, 2], num_classes=num_classes, in_chans=in_chans,
+        block_args=dict(sk_kwargs=sk_kwargs), **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model
+
+
+@register_model
+def sksresnet18(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a ResNet-18 model.
+    """
+    default_cfg = default_cfgs['skresnet18']
+    sk_kwargs = dict(
+        min_attn_channels=16,
+        attn_reduction=8,
+        split_input=True
+    )
+    model = ResNet(
+        SelectiveKernelBasic, [2, 2, 2, 2], num_classes=num_classes, in_chans=in_chans,
+        block_args=dict(sk_kwargs=sk_kwargs), zero_init_last_bn=False, **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model
+
+
+@register_model
+def skresnet26d(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a ResNet-26 model.
+    """
+    default_cfg = default_cfgs['skresnet26d']
+    sk_kwargs = dict(
+        keep_3x3=False,
+    )
+    model = ResNet(
+        SelectiveKernelBottleneck, [2, 2, 2, 2],  stem_width=32, stem_type='deep', avg_down=True,
+        num_classes=num_classes, in_chans=in_chans, block_args=dict(sk_kwargs=sk_kwargs), zero_init_last_bn=False
+        **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model
+
+
+@register_model
+def skresnet50(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a Select Kernel ResNet-50 model.
+    Based on config in "Compounding the Performance Improvements of Assembled Techniques in a
+    Convolutional Neural Network"
+    """
+    sk_kwargs = dict(
+        attn_reduction=2,
+    )
+    default_cfg = default_cfgs['skresnet50']
+    model = ResNet(
+        SelectiveKernelBottleneck, [3, 4, 6, 3], num_classes=num_classes, in_chans=in_chans,
+        block_args=dict(sk_kwargs=sk_kwargs), zero_init_last_bn=False, **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model
+
+
+@register_model
+def skresnet50d(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a Select Kernel ResNet-50-D model.
+    Based on config in "Compounding the Performance Improvements of Assembled Techniques in a
+    Convolutional Neural Network"
+    """
+    sk_kwargs = dict(
+        attn_reduction=2,
+    )
+    default_cfg = default_cfgs['skresnet50d']
+    model = ResNet(
+        SelectiveKernelBottleneck, [3, 4, 6, 3], stem_width=32, stem_type='deep', avg_down=True,
+        num_classes=num_classes, in_chans=in_chans, block_args=dict(sk_kwargs=sk_kwargs),
+        zero_init_last_bn=False, **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model
+
+
+@register_model
+def skresnext50_32x4d(pretrained=False, num_classes=1000, in_chans=3, **kwargs):
+    """Constructs a Select Kernel ResNeXt50-32x4d model. This should be equivalent to
+    the SKNet50 model in the Select Kernel Paper
+    """
+    default_cfg = default_cfgs['skresnext50_32x4d']
+    model = ResNet(
+        SelectiveKernelBottleneck, [3, 4, 6, 3], cardinality=32, base_width=4,
+        num_classes=num_classes, in_chans=in_chans, zero_init_last_bn=False, **kwargs)
+    model.default_cfg = default_cfg
+    if pretrained:
+        load_pretrained(model, default_cfg, num_classes, in_chans)
+    return model