Auto commit: 2024-12-12 22:00:01

models/superformer.py

@@ -735,12 +735,12 @@ class SuperFormer(nn.Module):
     """
 
     def __init__(self, img_size=12, patch_size=3, in_chans=1,
-                 embed_dim=48, depths=[6, 6, 6, 6], num_heads=[6, 6, 6, 6],
-                 window_size=6, mlp_ratio=4., qkv_bias=True, qk_scale=None,
+                 embed_dim=72, depths=[6, 6, 6, 6], num_heads=[6, 6, 6, 6],
+                 window_size=2, mlp_ratio=4., qkv_bias=True, qk_scale=None,
                  drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
                  norm_layer=nn.LayerNorm, ape=False, rpb=True ,patch_norm=True,
                  use_checkpoint=False, upscale=2, img_range=1., upsampler='', resi_connection='1conv',
-                 output_type = "residual",num_feat=64,**kwargs):
+                 output_type = "",num_feat=64,**kwargs):
         super(SuperFormer, self).__init__()
         num_in_ch = in_chans
         num_out_ch = in_chans
@@ -951,12 +951,26 @@ def forward(self, x):
                 res_deep = (res_deep_feat + res_deep_vol)/2
                 # Modify the upsampling to actually scale up the features
                 res = self.upsample_feat(res_deep)  # Make sure this actually upsamples
-                res = res + F.interpolate(x_first, scale_factor=self.upscale, mode='trilinear', align_corners=False)
+                x_first_upsampled = F.interpolate(x_first, 
+                                        size=(H*self.upscale, W*self.upscale, D*self.upscale), 
+                                        mode='trilinear', 
+                                        align_corners=False)
+                res_upsampled = F.interpolate(res, 
+                                        size=(H*self.upscale, W*self.upscale, D*self.upscale), 
+                                        mode='trilinear', 
+                                        align_corners=False)
+                res = res_upsampled + x_first_upsampled
             else:
-                res = self.conv_after_body(self.forward_features(x_first)) 
-                # Add upsampling here
-                res = F.interpolate(res, scale_factor=self.upscale, mode='trilinear', align_corners=False)
-                res = res + F.interpolate(x_first, scale_factor=self.upscale, mode='trilinear', align_corners=False)
+                res = self.conv_after_body(self.forward_features(x, x_first)) 
+                res = F.interpolate(res, 
+                          size=(H*self.upscale, W*self.upscale, D*self.upscale), 
+                          mode='trilinear', 
+                          align_corners=False)
+                x_first_upsampled = F.interpolate(x_first, 
+                                            size=(H*self.upscale, W*self.upscale, D*self.upscale), 
+                                            mode='trilinear', 
+                                            align_corners=False)
+                res = res + x_first_upsampled
             if self.output_type == 'residual':
                 x = x + self.conv_last(res)
             else:

requirements.txt

@@ -9,3 +9,4 @@ seaborn==0.13.2
 torch==2.3.0
 torchmetrics
 tqdm==4.66.4
+timm