#5 edit model name / fix model fc gender layer

Submission_codes/model.py

@@ -59,16 +59,16 @@ def forward(self, x):
         return x
 
 # Custom Model Template
-class Res183Ways(nn.Module):
+class Res18_3MultiLabel(nn.Module):
     def __init__(self, num_classes):
         super().__init__()
         self.res18 = Res18(num_classes)
-        self.res18.load_state_dict(torch.load("./model/res18/best.pth"))
+        self.res18.load_state_dict(torch.load("/opt/ml/workspace/code/model/Reres18/best.pth"))
         self.res18 = nn.Sequential(*list(self.res18.pretrain_model.children())[:-1])
 
         self.mask = nn.Linear(512, 3, bias=True)
         self.age = nn.Linear(512, 3, bias=True)
-        self.gender = nn.Linear(512, 3, bias=True)
+        self.gender = nn.Linear(512, 2, bias=True)
 
         def dfs_freeze(model):
             for name, child in model.named_children():
@@ -89,16 +89,16 @@ def forward(self, x):
         return {"mask":m, "age":a, "gender":s}
 
 # Custom Model Template
-class Res503Ways(nn.Module):
+class Res50_3MultiLabel(nn.Module):
     def __init__(self, num_classes):
         super().__init__()
         self.res50 = Res50(num_classes)
-        self.res50.load_state_dict(torch.load("./model/res50/best.pth"))
+        self.res50.load_state_dict(torch.load("/opt/ml/workspace/code/model/Reres50/best.pth"))
         self.res50 = nn.Sequential(*list(self.res50.pretrain_model.children())[:-1])
 
         self.mask = nn.Linear(2048, 3, bias=True)
         self.age = nn.Linear(2048, 3, bias=True)
-        self.gender = nn.Linear(2048, 3, bias=True)
+        self.gender = nn.Linear(2048, 2, bias=True)
 
         def dfs_freeze(model):
             for name, child in model.named_children():
@@ -117,3 +117,111 @@ def forward(self, x):
         a = self.age(x)
         s = self.gender(x)
         return {"mask":m, "age":a, "gender":s}
+
+
+
+# Custom Model Template
+class Res18_2MultiLabel(nn.Module):
+    def __init__(self, num_classes):
+        # super().__init__()
+        # self.res50 = Res50(num_classes)
+        # self.res50.load_state_dict(torch.load("./model/Reres50/best.pth"))
+        # self.res50 = nn.Sequential(*list(self.res50.pretrain_model.children())[:-1])
+        super().__init__()
+        self.res18 = Res18(num_classes)
+        self.res18.load_state_dict(torch.load("/opt/ml/workspace/code/model/Reres18/best.pth"))
+        self.res18 = nn.Sequential(*list(self.res18.pretrain_model.children())[:-1])
+
+        self.mask = nn.Linear(512, 3, bias=True)
+        self.mask.load_state_dict(torch.load("./resnext50_32x4dfc3ways_maskv3.pt"))
+
+        self.age = nn.Linear(524, 3, bias=True)
+        self.gender = nn.Linear(524, 2, bias=True)
+        def dfs_freeze(model):
+            for name, child in model.named_children():
+                for param in child.parameters():
+                    #print(param)
+                    param.requires_grad = False
+                    #print(param)
+                dfs_freeze(child)
+        dfs_freeze(self.res18)
+
+    def forward(self, x):
+        x = self.res18.forward(x)
+        x = torch.flatten(x, start_dim=1)
+
+        m = self.mask(x)
+
+        pred_mask = torch.argmax(m, dim=-1).cpu().numpy()
+        nx = []
+        base = torch.ones(12)
+        for i,k in enumerate(pred_mask):
+            v = int(k)
+            if v == 0: 
+                base = base * 0
+            elif v == 1:
+                base = base * 10
+            elif v == 2:
+                base = base * -10
+            tmp = torch.cat([x[i], self.base.to("cuda")])
+            nx.append(tmp)
+
+        nx = torch.stack(nx)
+        #print(nx.shape)
+        a = self.age(nx)
+        g = self.gender(nx)
+        return {"mask":m, "age":a, "gender":g}
+
+
+# Custom Model Template
+class Res50_M2MultiLabel(nn.Module):
+    def __init__(self, num_classes):
+        # super().__init__()
+        # self.res50 = Res50(num_classes)
+        # self.res50.load_state_dict(torch.load("./model/Reres50/best.pth"))
+        # self.res50 = nn.Sequential(*list(self.res50.pretrain_model.children())[:-1])
+        super().__init__()
+        self.res50 = Res50(num_classes)
+        self.res50.load_state_dict(torch.load("/opt/ml/workspace/code/model/Reres50/best.pth"))
+        self.res50 = nn.Sequential(*list(self.res50.pretrain_model.children())[:-1])
+
+        self.mask = nn.Linear(2048, 3, bias=True)
+        self.mask.load_state_dict(torch.load("./resnext50_32x4dfc3ways_maskv3.pt"))
+
+        self.age = nn.Linear(2060, 3, bias=True)
+        self.gender = nn.Linear(2060, 2, bias=True)
+
+        def dfs_freeze(model):
+            for name, child in model.named_children():
+                for param in child.parameters():
+                    #print(param)
+                    param.requires_grad = False
+                    #print(param)
+                dfs_freeze(child)
+        dfs_freeze(self.res50)
+
+    def forward(self, x):
+        x = self.res50.forward(x)
+        x = torch.flatten(x, start_dim=1)
+
+        m = self.mask(x)
+
+        pred_mask = torch.argmax(m, dim=-1).cpu().numpy()
+        nx = []
+        self.base = torch.ones(12)
+        for i,k in enumerate(pred_mask):
+            v = int(k)
+            if v == 0: 
+                self.base = self.base * 0
+            elif v == 1:
+                self.base = self.base * 10
+            elif v == 2:
+                self.base = self.base * -10
+            tmp = torch.cat([x[i], self.base.to("cuda")])
+            nx.append(tmp)
+
+        nx = torch.stack(nx)
+        #print(nx.shape)
+        a = self.age(nx)
+        g = self.gender(nx)
+        return {"mask":m, "age":a, "gender":g}