Remove unnecessary codes added in bc36c8d

vsrife/__init__.py

@@ -67,7 +67,7 @@ def RIFE(clip: vs.VideoNode, model_ver: float=3.5, scale: float=1.0, device_type
         from .model38.RIFE_HDv3 import Model
         model_dir = 'model38'
 
-    model = Model(device, fp16)
+    model = Model(device)
     model.load_model(os.path.join(os.path.dirname(__file__), model_dir), -1)
     model.eval()
     model.device()

vsrife/loss.py

@@ -16,16 +16,14 @@ def forward(self, flow, gt, loss_mask):
 
 
 class Ternary(nn.Module):
-    def __init__(self, device, fp16):
+    def __init__(self, device):
         super(Ternary, self).__init__()
         patch_size = 7
         out_channels = patch_size * patch_size
         self.w = np.eye(out_channels).reshape(
             (patch_size, patch_size, 1, out_channels))
         self.w = np.transpose(self.w, (3, 2, 0, 1))
         self.w = torch.tensor(self.w).float().to(device)
-        if fp16:
-            self.w = self.w.half()
 
     def transform(self, img):
         patches = F.conv2d(img, self.w, padding=3, bias=None)
@@ -56,15 +54,13 @@ def forward(self, img0, img1):
 
 
 class SOBEL(nn.Module):
-    def __init__(self, device, fp16):
+    def __init__(self, device):
         super(SOBEL, self).__init__()
         self.kernelX = torch.tensor([
             [1, 0, -1],
             [2, 0, -2],
             [1, 0, -1],
         ]).float()
-        if fp16:
-            self.kernelX = self.kernelX.half()
         self.kernelY = self.kernelX.clone().T
         self.kernelX = self.kernelX.unsqueeze(0).unsqueeze(0).to(device)
         self.kernelY = self.kernelY.unsqueeze(0).unsqueeze(0).to(device)

vsrife/model31/RIFE_HDv3.py

@@ -112,7 +112,7 @@ def forward(self, img0, img1, flow, c0, c1, flow_gt):
 
 
 class Model:
-    def __init__(self, device, fp16, local_rank=-1):
+    def __init__(self, device, local_rank=-1):
         self.torch_device = device
         self.flownet = IFNet(device)
         self.contextnet = ContextNet(device)
@@ -125,8 +125,8 @@ def __init__(self, device, fp16, local_rank=-1):
         self.schedulerG = optim.lr_scheduler.CyclicLR(
             self.optimG, base_lr=1e-6, max_lr=1e-3, step_size_up=8000, cycle_momentum=False)
         self.epe = EPE()
-        self.ter = Ternary(device, fp16)
-        self.sobel = SOBEL(device, fp16)
+        self.ter = Ternary(device)
+        self.sobel = SOBEL(device)
         if local_rank != -1:
             self.flownet = DDP(self.flownet, device_ids=[
                                local_rank], output_device=local_rank)

vsrife/model35/RIFE_HDv3.py

@@ -11,14 +11,14 @@
 from vsrife.loss import *
 
 class Model:
-    def __init__(self, device, fp16, local_rank=-1):
+    def __init__(self, device, local_rank=-1):
         self.torch_device = device
         self.flownet = IFNet(device)
         self.device()
         self.optimG = AdamW(self.flownet.parameters(), lr=1e-6, weight_decay=1e-4)
         self.epe = EPE()
         # self.vgg = VGGPerceptualLoss().to(device)
-        self.sobel = SOBEL(device, fp16)
+        self.sobel = SOBEL(device)
         if local_rank != -1:
             self.flownet = DDP(self.flownet, device_ids=[local_rank], output_device=local_rank)
 

vsrife/model38/RIFE_HDv3.py

@@ -11,14 +11,14 @@
 from vsrife.loss import *
 
 class Model:
-    def __init__(self, device, fp16, local_rank=-1):
+    def __init__(self, device, local_rank=-1):
         self.torch_device = device
         self.flownet = IFNet(device)
         self.device()
         self.optimG = AdamW(self.flownet.parameters(), lr=1e-6, weight_decay=1e-4)
         self.epe = EPE()
         # self.vgg = VGGPerceptualLoss().to(device)
-        self.sobel = SOBEL(device, fp16)
+        self.sobel = SOBEL(device)
         if local_rank != -1:
             self.flownet = DDP(self.flownet, device_ids=[local_rank], output_device=local_rank)