NSG improvements

- Just use resnet blocks for the multiplexer trunk of the generator
- Every block initializes itself, rather than everything at the end
- Cleans up some messy parts of the architecture, including unnecessary
  kernel sizes and places where BN is not used properly.

codes/models/archs/NestedSwitchGenerator.py

@@ -1,10 +1,11 @@
 import torch
 from torch import nn
-from models.archs.SwitchedResidualGenerator_arch import ConvBnLelu, create_sequential_growing_processing_block, MultiConvBlock, initialize_weights
+from models.archs.SwitchedResidualGenerator_arch import ConvBnLelu, MultiConvBlock, initialize_weights
 from switched_conv import BareConvSwitch, compute_attention_specificity
 from switched_conv_util import save_attention_to_image
 from functools import partial
 import torch.nn.functional as F
+from torchvision.models.resnet import BasicBlock, Bottleneck
 
 
 class Switch(nn.Module):
@@ -55,29 +56,22 @@ class Switch(nn.Module):
             [t.set_temperature(temp) for t in self.transforms]
 
 
-class ResidualBlock(nn.Module):
-    def __init__(self, filters):
-        super(ResidualBlock, self).__init__()
-        self.lelu1 = nn.LeakyReLU(negative_slope=.1)
-        self.bn1 = nn.BatchNorm2d(filters)
-        self.conv1 = nn.Conv2d(filters, filters, kernel_size=3, padding=1)
-        self.lelu2 = nn.LeakyReLU(negative_slope=.1)
-        self.bn2 = nn.BatchNorm2d(filters)
-        self.conv2 = nn.Conv2d(filters, filters, kernel_size=3, padding=1)
-
-    def forward(self, x):
-        x = self.conv1(self.lelu1(self.bn1(x)))
-        return self.conv2(self.lelu2(self.bn2(x)))
-
-
 # Convolutional image processing block that optionally reduces image size by a factor of 2 using stride and performs a
 # series of residual-block-like processing operations on it.
 class Processor(nn.Module):
     def __init__(self, base_filters, processing_depth, reduce=False):
         super(Processor, self).__init__()
-        self.output_filter_count = base_filters * 2 if reduce else base_filters
+        self.output_filter_count = base_filters * 2
-        self.initial = ConvBnLelu(base_filters, self.output_filter_count, kernel_size=1, stride=2 if reduce else 1)
+
-        self.res_blocks = nn.ModuleList([ResidualBlock(self.output_filter_count) for _ in range(processing_depth)])
+        # Downsample block used for bottleneck.
+        downsample = nn.Sequential(
+            nn.Conv2d(base_filters, self.output_filter_count, kernel_size=1, stride=2),
+            nn.BatchNorm2d(self.output_filter_count),
+        )
+        # Bottleneck block outputs the requested filter sizex4, but we only want x2.
+        self.initial = Bottleneck(base_filters, base_filters // 2, stride=2 if reduce else 1, downsample=downsample)
+
+        self.res_blocks = nn.ModuleList([BasicBlock(self.output_filter_count, self.output_filter_count) for _ in range(processing_depth)])
 
     def forward(self, x):
         x = self.initial(x)
@@ -89,14 +83,22 @@ class Processor(nn.Module):
 # Convolutional image processing block that constricts an input image with a large number of filters to a small number
 # of filters over a fixed number of layers.
 class Constrictor(nn.Module):
-    def __init__(self, filters, output_filters, use_bn=False):
+    def __init__(self, filters, output_filters):
         super(Constrictor, self).__init__()
         assert(filters > output_filters)
         gap = filters - output_filters
         gap_div_4 = int(gap / 4)
-        self.cbl1 = ConvBnLelu(filters, filters - (gap_div_4 * 2), bn=use_bn)
+        self.cbl1 = ConvBnLelu(filters, filters - (gap_div_4 * 2), kernel_size=1, bn=True)
-        self.cbl2 = ConvBnLelu(filters - (gap_div_4 * 2), filters - (gap_div_4 * 3), bn=use_bn)
+        self.cbl2 = ConvBnLelu(filters - (gap_div_4 * 2), filters - (gap_div_4 * 3), kernel_size=1, bn=True)
-        self.cbl3 = ConvBnLelu(filters - (gap_div_4 * 3), output_filters)
+        self.cbl3 = nn.Conv2d(filters - (gap_div_4 * 3), output_filters, kernel_size=1)
+
+        # Init params.
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='leaky_relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
 
     def forward(self, x):
         x = self.cbl1(x)
@@ -148,6 +150,24 @@ class NestedSwitchComputer(nn.Module):
         self.switch = RecursiveSwitchedTransform(transform_filters, filters, nesting_depth-1, transforms_at_leaf, trans_kernel_size, trans_num_layers-1, trans_scale_init, initial_temp=initial_temp, add_scalable_noise_to_transforms=add_scalable_noise_to_transforms)
         self.anneal = ConvBnLelu(transform_filters, transform_filters, kernel_size=1, bn=False)
 
+        # Init the parameters in the trunk.
+        for m in self.processing_trunk.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+        nn.init.kaiming_normal_(self.anneal.conv.weight, mode='fan_out', nonlinearity='leaky_relu')
+
+        # Zero-initialize the last BN in each residual branch,
+        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
+        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
+        for m in self.processing_trunk.modules():
+            if isinstance(m, Bottleneck):
+                nn.init.constant_(m.bn3.weight, 0)
+            elif isinstance(m, BasicBlock):
+                nn.init.constant_(m.bn2.weight, 0)
+
     def forward(self, x):
         trunk = []
         trunk_input = self.multiplexer_init_conv(x)
@@ -167,16 +187,17 @@ class NestedSwitchedGenerator(nn.Module):
                  trans_layers, transformation_filters, initial_temp=20, final_temperature_step=50000, heightened_temp_min=1,
                  heightened_final_step=50000, upsample_factor=1, add_scalable_noise_to_transforms=False):
         super(NestedSwitchedGenerator, self).__init__()
-        self.initial_conv = ConvBnLelu(3, transformation_filters, bn=False)
+        self.initial_conv = ConvBnLelu(3, transformation_filters, kernel_size=7, bn=False)
-        self.final_conv = ConvBnLelu(transformation_filters, 3, bn=False)
+        self.final_conv = ConvBnLelu(transformation_filters, 3, kernel_size=1, bn=False)
 
         switches = []
         for sw_reduce, sw_proc, trans_count, kernel, layers in zip(switch_reductions, switch_processing_layers, trans_counts, trans_kernel_sizes, trans_layers):
             switches.append(NestedSwitchComputer(transform_filters=transformation_filters, switch_base_filters=switch_filters, num_switch_processing_layers=sw_proc,
                                                  nesting_depth=sw_reduce, transforms_at_leaf=trans_count, trans_kernel_size=kernel, trans_num_layers=layers,
                                                  trans_scale_init=.2/len(switch_reductions), initial_temp=initial_temp, add_scalable_noise_to_transforms=add_scalable_noise_to_transforms))
-        initialize_weights(switches, 1)
         self.switches = nn.ModuleList(switches)
+        nn.init.kaiming_normal_(self.initial_conv.conv.weight, mode='fan_out', nonlinearity='leaky_relu')
+        nn.init.kaiming_normal_(self.final_conv.conv.weight, mode='fan_in', nonlinearity='leaky_relu')
 
         self.transformation_counts = trans_counts
         self.init_temperature = initial_temp
@@ -187,6 +208,7 @@ class NestedSwitchedGenerator(nn.Module):
         self.upsample_factor = upsample_factor
 
     def forward(self, x):
+        k = x
         # This network is entirely a "repair" network and operates on full-resolution images. Upsample first if that
         # is called for, then repair.
         if self.upsample_factor > 1:

codes/models/archs/SwitchedResidualGenerator_arch.py

@@ -44,6 +44,14 @@ class MultiConvBlock(nn.Module):
         self.scale = nn.Parameter(torch.full((1,), fill_value=scale_init))
         self.bias = nn.Parameter(torch.zeros(1))
 
+        # Init params.
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='leaky_relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
     def forward(self, x, noise=None):
         if noise is not None:
             noise = noise * self.noise_scale