Add new 3ph model without dropout

src/models.py

@@ -316,5 +316,64 @@ class triplePhotonNet_260529(nn.Module): ## adapted from doublePhotonNet_260507,
         
         flat_feat = c4.view(c4.size(0), -1)  # [B, 6272]
         
+        coords = self.fc(flat_feat)     # [B, 6]
+        return coords
+
+
+class triplePhotonNet_260611(nn.Module): ## adapted from triplePhotonNet_260529
+    def __init__(self):
+        super().__init__()
+        # Backbone: deeper for 9x9 input containing 3 photons
+        self.conv1 = nn.Conv2d(3, 32, kernel_size=3, padding=1) ### 9x9
+        self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1) ## 9x9
+        self.conv3 = nn.Conv2d(64, 128, kernel_size=3) ## 7x7
+        self.conv4 = nn.Conv2d(128, 128, kernel_size=3) ## 5x5
+        
+        # Spatial Attention Module
+        self.spatial_attn = nn.Sequential(
+            nn.Conv2d(128, 1, kernel_size=1),
+            nn.Sigmoid()
+        )
+        
+        self.fc = nn.Sequential(
+            nn.Linear(128 * 5 * 5, 512),
+            nn.ReLU(),
+            nn.Linear(512, 128),
+            nn.ReLU(),
+            nn.Linear(128, 6)
+        )
+        
+        self._init_weights()
+        self._init_coords()
+
+    def _init_coords(self):
+        # Create a coordinate grid; moved from dataset generation to model initialization for lower traffic and more flexibility
+        nSize = 9  # should match the input size of the model
+        x = np.linspace(-nSize/2. + 0.5, nSize/2. - 0.5, nSize)
+        y = np.linspace(-nSize/2. + 0.5, nSize/2. - 0.5, nSize)
+        x_grid, y_grid = np.meshgrid(x, y, indexing='ij')  # (nSize,nSize), (nSize,nSize)
+        self.x_grid = torch.tensor(np.expand_dims(x_grid, axis=0)).float().contiguous().to('cuda')  # (1, nSize, nSize)
+        self.y_grid = torch.tensor(np.expand_dims(y_grid, axis=0)).float().contiguous().to('cuda')  # (1, nSize, nSize)
+
+    def _init_weights(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, nn.Linear):
+                nn.init.xavier_uniform_(m.weight)
+                nn.init.zeros_(m.bias)
+    
+    def forward(self, x):
+        x = torch.cat((x, self.x_grid.expand(x.size(0), -1, -1, -1), self.y_grid.expand(x.size(0), -1, -1, -1)), dim=1)  # [B, 3, 9, 9]
+        c1 = F.relu(self.conv1(x))      # [B, 32, 9, 9]
+        c2 = F.relu(self.conv2(c1))     # [B, 64, 9, 9]
+        c3 = F.relu(self.conv3(c2))     # [B, 128, 7, 7]
+        c4 = F.relu(self.conv4(c3))     # [B, 128, 5, 5]
+        
+        attn = self.spatial_attn(c4)    # [B, 1, 5, 5]
+        c4 = c4 * attn                  # [B, 128, 5, 5]
+        
+        flat_feat = c4.view(c4.size(0), -1)  # [B, 3200]
+        
         coords = self.fc(flat_feat)     # [B, 6]
         return coords