diff --git a/.gitmodules b/.gitmodules
index bf1b51ab..3cd33570 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -4,3 +4,6 @@
 [submodule "codes/models/flownet2"]
 	path = codes/models/flownet2
 	url = https://github.com/NVIDIA/flownet2-pytorch.git
+[submodule "codes/switched_conv"]
+	path = codes/switched_conv
+	url = https://github.com/neonbjb/SwitchedConvolutions.git
diff --git a/codes/data_scripts/use_discriminator_as_filter.py b/codes/data_scripts/use_discriminator_as_filter.py
index 89df1344..078641a8 100644
--- a/codes/data_scripts/use_discriminator_as_filter.py
+++ b/codes/data_scripts/use_discriminator_as_filter.py
@@ -9,8 +9,8 @@ import options.options as option
 import utils.util as util
 from data.util import bgr2ycbcr
 import models.archs.SwitchedResidualGenerator_arch as srg
-from switched_conv_util import save_attention_to_image, save_attention_to_image_rgb
-from switched_conv import compute_attention_specificity
+from switched_conv.switched_conv_util import save_attention_to_image, save_attention_to_image_rgb
+from switched_conv.switched_conv import compute_attention_specificity
 from data import create_dataset, create_dataloader
 from models import create_model
 from tqdm import tqdm
diff --git a/codes/models/ExtensibleTrainer.py b/codes/models/ExtensibleTrainer.py
index 02e669b1..4f2b7636 100644
--- a/codes/models/ExtensibleTrainer.py
+++ b/codes/models/ExtensibleTrainer.py
@@ -109,7 +109,7 @@ class ExtensibleTrainer(BaseModel):
             if opt['dist']:
                 dnet = DistributedDataParallel(anet,
                                                device_ids=[torch.cuda.current_device()],
-                                               find_unused_parameters=True)
+                                               find_unused_parameters=False)
             else:
                 dnet = DataParallel(anet)
             if self.is_train:
diff --git a/codes/models/archs/ProgressiveSrg_arch.py b/codes/models/archs/ProgressiveSrg_arch.py
index 59988dda..f2ff9441 100644
--- a/codes/models/archs/ProgressiveSrg_arch.py
+++ b/codes/models/archs/ProgressiveSrg_arch.py
@@ -1,8 +1,8 @@
 import models.archs.SwitchedResidualGenerator_arch as srg
 import torch
 import torch.nn as nn
-from switched_conv_util import save_attention_to_image
-from switched_conv import compute_attention_specificity
+from switched_conv.switched_conv_util import save_attention_to_image
+from switched_conv.switched_conv import compute_attention_specificity
 from models.archs.arch_util import ConvGnLelu, ExpansionBlock, MultiConvBlock
 import functools
 import torch.nn.functional as F
diff --git a/codes/models/archs/SPSR_arch.py b/codes/models/archs/SPSR_arch.py
index 89ddb060..26240514 100644
--- a/codes/models/archs/SPSR_arch.py
+++ b/codes/models/archs/SPSR_arch.py
@@ -12,8 +12,8 @@ from models.archs import SPSR_util as B
 from models.archs.SwitchedResidualGenerator_arch import ConfigurableSwitchComputer, ReferenceImageBranch, \
     QueryKeyMultiplexer, QueryKeyPyramidMultiplexer
 from models.archs.arch_util import ConvGnLelu, UpconvBlock, MultiConvBlock, ReferenceJoinBlock
-from switched_conv import compute_attention_specificity
-from switched_conv_util import save_attention_to_image_rgb
+from switched_conv.switched_conv import compute_attention_specificity
+from switched_conv.switched_conv_util import save_attention_to_image_rgb
 from .RRDBNet_arch import RRDB
 
 
diff --git a/codes/models/archs/StructuredSwitchedGenerator.py b/codes/models/archs/StructuredSwitchedGenerator.py
index d8e7d638..6533719a 100644
--- a/codes/models/archs/StructuredSwitchedGenerator.py
+++ b/codes/models/archs/StructuredSwitchedGenerator.py
@@ -6,8 +6,8 @@ from models.archs.SPSR_arch import ImageGradientNoPadding
 from torch import nn
 import torch
 import torch.nn.functional as F
-from switched_conv_util import save_attention_to_image_rgb
-from switched_conv import compute_attention_specificity
+from switched_conv.switched_conv_util import save_attention_to_image_rgb
+from switched_conv.switched_conv import compute_attention_specificity
 import os
 import torchvision
 from torch.utils.checkpoint import checkpoint
diff --git a/codes/models/archs/SwitchedResidualGenerator_arch.py b/codes/models/archs/SwitchedResidualGenerator_arch.py
index 99145063..e2345912 100644
--- a/codes/models/archs/SwitchedResidualGenerator_arch.py
+++ b/codes/models/archs/SwitchedResidualGenerator_arch.py
@@ -1,11 +1,11 @@
 import torch
 from torch import nn
-from switched_conv import BareConvSwitch, compute_attention_specificity, AttentionNorm
+from switched_conv.switched_conv import BareConvSwitch, compute_attention_specificity, AttentionNorm
 import torch.nn.functional as F
 import functools
 from collections import OrderedDict
 from models.archs.arch_util import ConvBnLelu, ConvGnSilu, ExpansionBlock, ExpansionBlock2, ConvGnLelu, MultiConvBlock, SiLU
-from switched_conv_util import save_attention_to_image_rgb
+from switched_conv.switched_conv_util import save_attention_to_image_rgb
 import os
 from models.archs.spinenet_arch import SpineNet
 
diff --git a/codes/switched_conv b/codes/switched_conv
new file mode 160000
index 00000000..b69d0317
--- /dev/null
+++ b/codes/switched_conv
@@ -0,0 +1 @@
+Subproject commit b69d031766e042d40ecd7dc4986422e77ab278cf
diff --git a/codes/test.py b/codes/test.py
index 2c272d1f..ffd8fa7e 100644
--- a/codes/test.py
+++ b/codes/test.py
@@ -9,8 +9,8 @@ import options.options as option
 import utils.util as util
 from data.util import bgr2ycbcr
 import models.archs.SwitchedResidualGenerator_arch as srg
-from switched_conv_util import save_attention_to_image, save_attention_to_image_rgb
-from switched_conv import compute_attention_specificity
+from switched_conv.switched_conv_util import save_attention_to_image, save_attention_to_image_rgb
+from switched_conv.switched_conv import compute_attention_specificity
 from data import create_dataset, create_dataloader
 from models import create_model
 from tqdm import tqdm