From 70d781ec03e7975646daa7785e2f954c8f9aabe0 Mon Sep 17 00:00:00 2001
From: fumiama <s862105088@gmail.com>
Date: Tue, 1 Feb 2022 16:56:21 +0800
Subject: [PATCH] =?UTF-8?q?perf:=20=E4=BC=98=E5=8C=96=E5=86=85=E5=AD=98?=
 =?UTF-8?q?=E5=8D=A0=E7=94=A8?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 Real-CUGAN/upcunet_v3.py | 230 ++++++++++++++++++++++++---------------
 1 file changed, 143 insertions(+), 87 deletions(-)

diff --git a/Real-CUGAN/upcunet_v3.py b/Real-CUGAN/upcunet_v3.py
index 3ed5b5b..7218428 100644
--- a/Real-CUGAN/upcunet_v3.py
+++ b/Real-CUGAN/upcunet_v3.py
@@ -73,6 +73,7 @@ def __init__(self, in_channels, out_channels, deconv):
 
     def forward(self, x):
         x1 = self.conv1(x)
+        del x
         x2 = self.conv1_down(x1)
         x2 = F.leaky_relu(x2, 0.1, inplace=True)
         x2 = self.conv2(x2)
@@ -81,12 +82,15 @@ def forward(self, x):
 
         x1 = F.pad(x1, (-4, -4, -4, -4))
         x3 = self.conv3(x1 + x2)
+        del x1, x2
         x3 = F.leaky_relu(x3, 0.1, inplace=True)
         z = self.conv_bottom(x3)
+        del x3
         return z
 
     def forward_a(self, x):
         x1 = self.conv1(x)
+        del x
         x2 = self.conv1_down(x1)
         x2 = F.leaky_relu(x2, 0.1, inplace=True)
         x2 = self.conv2.conv(x2)
@@ -98,8 +102,10 @@ def forward_b(self, x1,x2):
 
         x1 = F.pad(x1, (-4, -4, -4, -4))
         x3 = self.conv3(x1 + x2)
+        del x1, x2
         x3 = F.leaky_relu(x3, 0.1, inplace=True)
         z = self.conv_bottom(x3)
+        del x3
         return z
 class UNet1x3(nn.Module):
     def __init__(self, in_channels, out_channels, deconv):
@@ -125,6 +131,7 @@ def __init__(self, in_channels, out_channels, deconv):
 
     def forward(self, x):
         x1 = self.conv1(x)
+        del x
         x2 = self.conv1_down(x1)
         x2 = F.leaky_relu(x2, 0.1, inplace=True)
         x2 = self.conv2(x2)
@@ -133,12 +140,15 @@ def forward(self, x):
 
         x1 = F.pad(x1, (-4, -4, -4, -4))
         x3 = self.conv3(x1 + x2)
+        del x1, x2
         x3 = F.leaky_relu(x3, 0.1, inplace=True)
         z = self.conv_bottom(x3)
+        del x3
         return z
 
     def forward_a(self, x):
         x1 = self.conv1(x)
+        del x
         x2 = self.conv1_down(x1)
         x2 = F.leaky_relu(x2, 0.1, inplace=True)
         x2 = self.conv2.conv(x2)
@@ -150,8 +160,10 @@ def forward_b(self, x1,x2):
 
         x1 = F.pad(x1, (-4, -4, -4, -4))
         x3 = self.conv3(x1 + x2)
+        del x1, x2
         x3 = F.leaky_relu(x3, 0.1, inplace=True)
         z = self.conv_bottom(x3)
+        del x3
         return z
 class UNet2(nn.Module):
     def __init__(self, in_channels, out_channels, deconv):
@@ -194,14 +206,17 @@ def forward(self, x):
 
         x2 = F.pad(x2, (-4, -4, -4, -4))
         x4 = self.conv4(x2 + x3)
+        del x2, x3
         x4 = self.conv4_up(x4)
         x4 = F.leaky_relu(x4, 0.1, inplace=True)
 
         x1 = F.pad(x1, (-16, -16, -16, -16))
         x5 = self.conv5(x1 + x4)
+        del x1, x4
         x5 = F.leaky_relu(x5, 0.1, inplace=True)
 
         z = self.conv_bottom(x5)
+        del x5
         return z
 
     def forward_a(self, x):#conv234结尾有se
@@ -223,6 +238,7 @@ def forward_c(self, x2,x3):  # conv234结尾有se
 
         x2 = F.pad(x2, (-4, -4, -4, -4))
         x4 = self.conv4.conv(x2 + x3)
+        del x2, x3
         return x4
 
     def forward_d(self, x1,x4):  # conv234结尾有se
@@ -231,9 +247,11 @@ def forward_d(self, x1,x4):  # conv234结尾有se
 
         x1 = F.pad(x1, (-16, -16, -16, -16))
         x5 = self.conv5(x1 + x4)
+        del x1, x4
         x5 = F.leaky_relu(x5, 0.1, inplace=True)
 
         z = self.conv_bottom(x5)
+        del x5
         return z
 class UpCunet2x(nn.Module):#完美tile，全程无损
     def __init__(self, in_channels=3, out_channels=3):
@@ -250,7 +268,9 @@ def forward(self, x,tile_mode):#1.7G
             x = self.unet1.forward(x)
             x0 = self.unet2.forward(x)
             x1 = F.pad(x, (-20, -20, -20, -20))
+            del x
             x = torch.add(x0, x1)
+            del x0, x1
             if (w0 != pw or h0 != ph): x = x[:, :, :h0 * 2, :w0 * 2]
             return x
         elif(tile_mode==1):# 对长边减半
@@ -261,19 +281,20 @@ def forward(self, x,tile_mode):#1.7G
                 crop_size_h=((h0-1)//4*4+4)//2#减半后能被2整除，所以要先被4整除
                 crop_size_w=(w0-1)//2*2+2#能被2整除
             crop_size=(crop_size_h,crop_size_w)#6.6G
-        elif(tile_mode==2):#hw都减半
-            crop_size=(((h0-1)//4*4+4)//2,((w0-1)//4*4+4)//2)#5.6G
-        elif(tile_mode==3):#hw都三分之一
-            crop_size=(((h0-1)//6*6+6)//3,((w0-1)//6*6+6)//3)#4.2G
-        elif(tile_mode==4):#hw都四分之一
-            crop_size=(((h0-1)//8*8+8)//4,((w0-1)//8*8+8)//4)#3.7G
+            #hw都减半 5.6G
+            #hw都三分之一 4.2G
+            #hw都四分之一 3.7G
+        else:
+            s = tile_mode*2
+            crop_size=(((h0-1)//s*s+s)//tile_mode,((w0-1)//s*s+s)//tile_mode)
         ph = ((h0 - 1) // crop_size[0] + 1) * crop_size[0]
         pw = ((w0 - 1) // crop_size[1] + 1) * crop_size[1]
         x=F.pad(x,(18,18+pw-w0,18,18+ph-h0),'reflect')
         n,c,h,w=x.shape
-        se_mean0=torch.zeros((n,64,1,1)).to(x.device)
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        xdev = x.device
+        se_mean0=torch.zeros((n,64,1,1)).to(xdev)
+        ishalf = "Half" in x.type()
+        if ishalf: se_mean0=se_mean0.half()
         n_patch=0
         tmp_dict={}
         opt_res_dict={}
@@ -283,77 +304,89 @@ def forward(self, x,tile_mode):#1.7G
                 x_crop=x[:,:,i:i+crop_size[0]+36,j:j+crop_size[1]+36]
                 n,c1,h1,w1=x_crop.shape
                 tmp0,x_crop = self.unet1.forward_a(x_crop)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(x_crop.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(x_crop, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 n_patch+=1
                 tmp_dict[i][j]=(tmp0,x_crop)
+                del tmp0, x_crop
+        del x
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 128, 1, 1)).to(x.device)#64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 128, 1, 1)).to(xdev)#64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-36,crop_size[0]):
             for j in range(0,w-36,crop_size[1]):
-                tmp0, x_crop=tmp_dict[i][j]
+                tmp0, x_crop = tmp_dict[i][j]
+                del tmp_dict[i][j]
                 x_crop=self.unet1.conv2.seblock.forward_mean(x_crop,se_mean0)
                 opt_unet1=self.unet1.forward_b(tmp0,x_crop)
+                del tmp0, x_crop
                 tmp_x1,tmp_x2 = self.unet2.forward_a(opt_unet1)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x2.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x2, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2)
+                del opt_unet1, tmp_x1, tmp_x2
         se_mean1/=n_patch
-        se_mean0 = torch.zeros((n, 128, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        se_mean0 = torch.zeros((n, 128, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean0=se_mean0.half()
         for i in range(0,h-36,crop_size[0]):
             for j in range(0,w-36,crop_size[1]):
                 opt_unet1,tmp_x1, tmp_x2=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x2=self.unet2.conv2.seblock.forward_mean(tmp_x2,se_mean1)
                 tmp_x3=self.unet2.forward_b(tmp_x2)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x3.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x3, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2,tmp_x3)
+                del opt_unet1, tmp_x1, tmp_x2, tmp_x3
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 64, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 64, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-36,crop_size[0]):
             for j in range(0,w-36,crop_size[1]):
                 opt_unet1,tmp_x1, tmp_x2,tmp_x3=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x3=self.unet2.conv3.seblock.forward_mean(tmp_x3,se_mean0)
                 tmp_x4=self.unet2.forward_c(tmp_x2,tmp_x3)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                del tmp_x2, tmp_x3
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x4.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x4, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x4)
+                del opt_unet1, tmp_x1, tmp_x4
         se_mean1/=n_patch
         for i in range(0,h-36,crop_size[0]):
             opt_res_dict[i]={}
             for j in range(0,w-36,crop_size[1]):
                 opt_unet1,tmp_x1, tmp_x4=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x4=self.unet2.conv4.seblock.forward_mean(tmp_x4,se_mean1)
                 x0=self.unet2.forward_d(tmp_x1,tmp_x4)
+                del tmp_x1, tmp_x4
                 x1 = F.pad(opt_unet1,(-20,-20,-20,-20))
+                del opt_unet1
                 x_crop = torch.add(x0, x1)#x0是unet2的最终输出
+                del x0, x1
                 opt_res_dict[i][j] = x_crop
+                del x_crop
         del tmp_dict
         torch.cuda.empty_cache()
-        res = torch.zeros((n, c, h * 2 - 72, w * 2 - 72)).to(x.device)
-        if ("Half" in x.type()):
-            res=res.half()
+        res = torch.zeros((n, c, h * 2 - 72, w * 2 - 72)).to(xdev)
+        if ishalf: res=res.half()
         for i in range(0,h-36,crop_size[0]):
             for j in range(0,w-36,crop_size[1]):
                 res[:, :, i * 2:i * 2 + h1 * 2 - 72, j * 2:j * 2 + w1 * 2 - 72]=opt_res_dict[i][j]
+                del opt_res_dict[i][j]
         del opt_res_dict
         torch.cuda.empty_cache()
         if(w0!=pw or h0!=ph):res=res[:,:,:h0*2,:w0*2]
@@ -374,6 +407,7 @@ def forward(self, x,tile_mode):#1.7G
             x0 = self.unet2.forward(x)
             x1 = F.pad(x, (-20, -20, -20, -20))
             x = torch.add(x0, x1)
+            del x0, x1
             if (w0 != pw or h0 != ph): x = x[:, :, :h0 * 3, :w0 * 3]
             return x
         elif(tile_mode==1):# 对长边减半
@@ -384,19 +418,17 @@ def forward(self, x,tile_mode):#1.7G
                 crop_size_h=((h0-1)//8*8+8)//2#减半后能被4整除，所以要先被8整除
                 crop_size_w=(w0-1)//4*4+4#能被4整除
             crop_size=(crop_size_h,crop_size_w)#6.6G
-        elif(tile_mode==2):#hw都减半
-            crop_size=(((h0-1)//8*8+8)//2,((w0-1)//8*8+8)//2)#5.6G
-        elif(tile_mode==3):#hw都三分之一
-            crop_size=(((h0-1)//12*12+12)//3,((w0-1)//12*12+12)//3)#4.2G
-        elif(tile_mode==4):#hw都四分之一
-            crop_size=(((h0-1)//16*16+16)//4,((w0-1)//16*16+16)//4)#3.7G
+        else:
+            s = tile_mode*2
+            crop_size=(((h0-1)//s*s+s)//tile_mode,((w0-1)//s*s+s)//tile_mode)
         ph = ((h0 - 1) // crop_size[0] + 1) * crop_size[0]
         pw = ((w0 - 1) // crop_size[1] + 1) * crop_size[1]
         x=F.pad(x,(14,14+pw-w0,14,14+ph-h0),'reflect')
         n,c,h,w=x.shape
-        se_mean0=torch.zeros((n,64,1,1)).to(x.device)
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        xdev = x.device
+        se_mean0=torch.zeros((n,64,1,1)).to(xdev)
+        ishalf = "Half" in x.type()
+        if ishalf: se_mean0=se_mean0.half()
         n_patch=0
         tmp_dict={}
         opt_res_dict={}
@@ -406,77 +438,89 @@ def forward(self, x,tile_mode):#1.7G
                 x_crop=x[:,:,i:i+crop_size[0]+28,j:j+crop_size[1]+28]
                 n,c1,h1,w1=x_crop.shape
                 tmp0,x_crop = self.unet1.forward_a(x_crop)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(x_crop.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(x_crop, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 n_patch+=1
                 tmp_dict[i][j]=(tmp0,x_crop)
+                del tmp0, x_crop
+        del x
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 128, 1, 1)).to(x.device)#64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 128, 1, 1)).to(xdev)#64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-28,crop_size[0]):
             for j in range(0,w-28,crop_size[1]):
                 tmp0, x_crop=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 x_crop=self.unet1.conv2.seblock.forward_mean(x_crop,se_mean0)
                 opt_unet1=self.unet1.forward_b(tmp0,x_crop)
+                del tmp0, x_crop
                 tmp_x1,tmp_x2 = self.unet2.forward_a(opt_unet1)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x2.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x2, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2)
+                del opt_unet1, tmp_x1, tmp_x2
         se_mean1/=n_patch
-        se_mean0 = torch.zeros((n, 128, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        se_mean0 = torch.zeros((n, 128, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean0=se_mean0.half()
         for i in range(0,h-28,crop_size[0]):
             for j in range(0,w-28,crop_size[1]):
-                opt_unet1,tmp_x1, tmp_x2=tmp_dict[i][j]
+                opt_unet1, tmp_x1, tmp_x2=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x2=self.unet2.conv2.seblock.forward_mean(tmp_x2,se_mean1)
                 tmp_x3=self.unet2.forward_b(tmp_x2)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x3.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x3, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2,tmp_x3)
+                del opt_unet1, tmp_x1, tmp_x2, tmp_x3
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 64, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 64, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-28,crop_size[0]):
             for j in range(0,w-28,crop_size[1]):
-                opt_unet1,tmp_x1, tmp_x2,tmp_x3=tmp_dict[i][j]
+                opt_unet1, tmp_x1, tmp_x2, tmp_x3=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x3=self.unet2.conv3.seblock.forward_mean(tmp_x3,se_mean0)
                 tmp_x4=self.unet2.forward_c(tmp_x2,tmp_x3)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                del tmp_x2, tmp_x3
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x4.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x4, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x4)
+                del opt_unet1, tmp_x1, tmp_x4
         se_mean1/=n_patch
         for i in range(0,h-28,crop_size[0]):
             opt_res_dict[i]={}
             for j in range(0,w-28,crop_size[1]):
-                opt_unet1,tmp_x1, tmp_x4=tmp_dict[i][j]
+                opt_unet1, tmp_x1, tmp_x4=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x4=self.unet2.conv4.seblock.forward_mean(tmp_x4,se_mean1)
                 x0=self.unet2.forward_d(tmp_x1,tmp_x4)
+                del tmp_x1, tmp_x4
                 x1 = F.pad(opt_unet1,(-20,-20,-20,-20))
+                del opt_unet1
                 x_crop = torch.add(x0, x1)#x0是unet2的最终输出
+                del x0, x1
                 opt_res_dict[i][j]=x_crop#
+                del x_crop
         del tmp_dict
         torch.cuda.empty_cache()
-        res = torch.zeros((n, c, h * 3 - 84, w * 3 - 84)).to(x.device)
-        if ("Half" in x.type()):
-            res=res.half()
+        res = torch.zeros((n, c, h * 3 - 84, w * 3 - 84)).to(xdev)
+        if ishalf: res=res.half()
         for i in range(0,h-28,crop_size[0]):
             for j in range(0,w-28,crop_size[1]):
-                res[:, :, i*3:i*3 + h1*3-84, j*3:j*3 + w1*3-84] =opt_res_dict[i][j]
+                res[:, :, i*3:i*3 + h1*3-84, j*3:j*3 + w1*3-84] = opt_res_dict[i][j]
+                del opt_res_dict[i][j]
         del opt_res_dict
         torch.cuda.empty_cache()
         if(w0!=pw or h0!=ph):res=res[:,:,:h0*3,:w0*3]
@@ -499,11 +543,13 @@ def forward(self, x,tile_mode):
             x0 = self.unet2.forward(x)
             x1 = F.pad(x, (-20, -20, -20, -20))
             x = torch.add(x0, x1)
+            del x0, x1
             x=self.conv_final(x)
             x=F.pad(x,(-1,-1,-1,-1))
             x=self.ps(x)
             if (w0 != pw or h0 != ph): x = x[:, :, :h0 * 4, :w0 * 4]
             x+=F.interpolate(x00, scale_factor=4, mode='nearest')
+            del x00
             return x
         elif(tile_mode==1):# 对长边减半
             if(w0>=h0):
@@ -513,19 +559,17 @@ def forward(self, x,tile_mode):
                 crop_size_h=((h0-1)//4*4+4)//2#减半后能被2整除，所以要先被4整除
                 crop_size_w=(w0-1)//2*2+2#能被2整除
             crop_size=(crop_size_h,crop_size_w)#6.6G
-        elif(tile_mode==2):#hw都减半
-            crop_size=(((h0-1)//4*4+4)//2,((w0-1)//4*4+4)//2)#5.6G
-        elif(tile_mode==3):#hw都三分之一
-            crop_size=(((h0-1)//6*6+6)//3,((w0-1)//6*6+6)//3)#4.1G
-        elif(tile_mode==4):#hw都四分之一
-            crop_size=(((h0-1)//8*8+8)//4,((w0-1)//8*8+8)//4)#3.7G
+        else:
+            s = tile_mode*2
+            crop_size=(((h0-1)//s*s+s)//tile_mode,((w0-1)//s*s+s)//tile_mode)
         ph = ((h0 - 1) // crop_size[0] + 1) * crop_size[0]
         pw = ((w0 - 1) // crop_size[1] + 1) * crop_size[1]
         x=F.pad(x,(19,19+pw-w0,19,19+ph-h0),'reflect')
         n,c,h,w=x.shape
-        se_mean0=torch.zeros((n,64,1,1)).to(x.device)
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        xdev = x.device
+        se_mean0=torch.zeros((n,64,1,1)).to(xdev)
+        ishalf = "Half" in x.type()
+        if ishalf: se_mean0=se_mean0.half()
         n_patch=0
         tmp_dict={}
         opt_res_dict={}
@@ -535,85 +579,96 @@ def forward(self, x,tile_mode):
                 x_crop=x[:,:,i:i+crop_size[0]+38,j:j+crop_size[1]+38]
                 n,c1,h1,w1=x_crop.shape
                 tmp0,x_crop = self.unet1.forward_a(x_crop)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(x_crop.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(x_crop, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 n_patch+=1
                 tmp_dict[i][j]=(tmp0,x_crop)
+                del tmp0, x_crop
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 128, 1, 1)).to(x.device)#64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 128, 1, 1)).to(xdev)#64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-38,crop_size[0]):
             for j in range(0,w-38,crop_size[1]):
                 tmp0, x_crop=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 x_crop=self.unet1.conv2.seblock.forward_mean(x_crop,se_mean0)
                 opt_unet1=self.unet1.forward_b(tmp0,x_crop)
+                del tmp0, x_crop
                 tmp_x1,tmp_x2 = self.unet2.forward_a(opt_unet1)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x2.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x2, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2)
+                del opt_unet1, tmp_x1, tmp_x2
         se_mean1/=n_patch
-        se_mean0 = torch.zeros((n, 128, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean0=se_mean0.half()
+        se_mean0 = torch.zeros((n, 128, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean0=se_mean0.half()
         for i in range(0,h-38,crop_size[0]):
             for j in range(0,w-38,crop_size[1]):
                 opt_unet1,tmp_x1, tmp_x2=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x2=self.unet2.conv2.seblock.forward_mean(tmp_x2,se_mean1)
                 tmp_x3=self.unet2.forward_b(tmp_x2)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x3.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x3, dim=(2, 3),keepdim=True)
                 se_mean0+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x2,tmp_x3)
+                del opt_unet1, tmp_x1, tmp_x2, tmp_x3
         se_mean0/=n_patch
-        se_mean1 = torch.zeros((n, 64, 1, 1)).to(x.device)  # 64#128#128#64
-        if ("Half" in x.type()):
-            se_mean1=se_mean1.half()
+        se_mean1 = torch.zeros((n, 64, 1, 1)).to(xdev)  # 64#128#128#64
+        if ishalf: se_mean1=se_mean1.half()
         for i in range(0,h-38,crop_size[0]):
             for j in range(0,w-38,crop_size[1]):
-                opt_unet1,tmp_x1, tmp_x2,tmp_x3=tmp_dict[i][j]
+                opt_unet1, tmp_x1, tmp_x2, tmp_x3=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x3=self.unet2.conv3.seblock.forward_mean(tmp_x3,se_mean0)
                 tmp_x4=self.unet2.forward_c(tmp_x2,tmp_x3)
-                if ("Half" in x.type()):  # torch.HalfTensor/torch.cuda.HalfTensor
+                del tmp_x2, tmp_x3
+                if ishalf:  # torch.HalfTensor/torch.cuda.HalfTensor
                     tmp_se_mean = torch.mean(tmp_x4.float(), dim=(2, 3),keepdim=True).half()
                 else:
                     tmp_se_mean = torch.mean(tmp_x4, dim=(2, 3),keepdim=True)
                 se_mean1+=tmp_se_mean
                 tmp_dict[i][j]=(opt_unet1,tmp_x1,tmp_x4)
+                del opt_unet1, tmp_x1, tmp_x4
         se_mean1/=n_patch
         for i in range(0,h-38,crop_size[0]):
             opt_res_dict[i]={}
             for j in range(0,w-38,crop_size[1]):
                 opt_unet1,tmp_x1, tmp_x4=tmp_dict[i][j]
+                del tmp_dict[i][j]
                 tmp_x4=self.unet2.conv4.seblock.forward_mean(tmp_x4,se_mean1)
                 x0=self.unet2.forward_d(tmp_x1,tmp_x4)
+                del tmp_x1, tmp_x4
                 x1 = F.pad(opt_unet1,(-20,-20,-20,-20))
+                del opt_unet1
                 x_crop = torch.add(x0, x1)#x0是unet2的最终输出
                 x_crop=self.conv_final(x_crop)
                 x_crop = F.pad(x_crop, (-1, -1, -1, -1))
                 x_crop=self.ps(x_crop)
                 opt_res_dict[i][j] = x_crop
+                del x_crop
         del tmp_dict
         torch.cuda.empty_cache()
-        res = torch.zeros((n, c, h * 4 - 152, w * 4 - 152)).to(x.device)
-        if ("Half" in x.type()):
-            res=res.half()
+        res = torch.zeros((n, c, h * 4 - 152, w * 4 - 152)).to(xdev)
+        if ishalf: res=res.half()
         for i in range(0,h-38,crop_size[0]):
             for j in range(0,w-38,crop_size[1]):
                 # print(opt_res_dict[i][j].shape,res[:, :, i * 4:i * 4 + h1 * 4 - 144, j * 4:j * 4 + w1 * 4 - 144].shape)
                 res[:, :, i * 4:i * 4 + h1 * 4 - 152, j * 4:j * 4 + w1 * 4 - 152]=opt_res_dict[i][j]
+                del opt_res_dict[i][j]
         del opt_res_dict
         torch.cuda.empty_cache()
         if(w0!=pw or h0!=ph):res=res[:,:,:h0*4,:w0*4]
         res += F.interpolate(x00, scale_factor=4, mode='nearest')
+        del x00
         return res#
 class RealWaifuUpScaler(object):
     def __init__(self,scale,weight_path,half,device):
@@ -638,12 +693,13 @@ def __call__(self, frame,tile_mode):
         with torch.no_grad():
             tensor = self.np2tensor(frame)
             result = self.tensor2np(self.model(tensor,tile_mode))
+            del tensor
         return result
 
 if __name__ == "__main__":
     ###########inference_img
-    import time, cv2,sys
-    from time import time as ttime
+    import cv2, sys
+    from time import time
     for weight_path,scale in [("weights_v3/up2x-latest-denoise3x.pth",2),("weights_v3/up3x-latest-denoise3x.pth",3),("weights_v3/up4x-latest-denoise3x.pth",4)]:
         for tile_mode in [0,1,2,3,4]:
             upscaler2x = RealWaifuUpScaler(scale, weight_path, half=True, device="cuda:0")
@@ -656,17 +712,17 @@ def __call__(self, frame,tile_mode):
                 inp_path = os.path.join(input_dir, name)
                 suffix = tmp[-1]
                 prefix = ".".join(tmp[:-1])
-                tmp_path = os.path.join(root_path, "tmp", "%s.%s" % (int(time.time() * 1000000), suffix))
+                tmp_path = os.path.join(root_path, "tmp", "%s.%s" % (int(time() * 1000000), suffix))
                 print(inp_path,tmp_path)
                 #支持中文路径
                 # os.link(inp_path, tmp_path)#win用硬链接
                 os.symlink(inp_path, tmp_path)#linux用软链接
                 frame = cv2.imread(tmp_path)[:, :, [2, 1, 0]]
-                t0 = ttime()
+                t0 = time()
                 result = upscaler2x(frame, tile_mode=tile_mode)[:, :, ::-1]
-                t1 = ttime()
+                t1 = time()
                 print(prefix, "done", t1 - t0)
-                tmp_opt_path = os.path.join(root_path, "tmp", "%s.%s" % (int(time.time() * 1000000), suffix))
+                tmp_opt_path = os.path.join(root_path, "tmp", "%s.%s" % (int(time() * 1000000), suffix))
                 cv2.imwrite(tmp_opt_path, result)
                 n=0
                 while(1):