Implement global average pooling for controlnet.

comfy/sd.py

@@ -621,7 +621,7 @@ def broadcast_image_to(tensor, target_batch_size, batched_number):
         return torch.cat([tensor] * batched_number, dim=0)
 
 class ControlNet:
-    def __init__(self, control_model, device=None):
+    def __init__(self, control_model, global_average_pooling=False, device=None):
         self.control_model = control_model
         self.cond_hint_original = None
         self.cond_hint = None
@@ -630,6 +630,7 @@ class ControlNet:
             device = model_management.get_torch_device()
         self.device = device
         self.previous_controlnet = None
+        self.global_average_pooling = global_average_pooling
 
     def get_control(self, x_noisy, t, cond_txt, batched_number):
         control_prev = None
@@ -665,6 +666,9 @@ class ControlNet:
                 key = 'output'
                 index = i
             x = control[i]
+            if self.global_average_pooling:
+                x = torch.mean(x, dim=(2, 3), keepdim=True).repeat(1, 1, x.shape[2], x.shape[3])
+
             x *= self.strength
             if x.dtype != output_dtype and not autocast_enabled:
                 x = x.to(output_dtype)
@@ -695,7 +699,7 @@ class ControlNet:
             self.cond_hint = None
 
     def copy(self):
-        c = ControlNet(self.control_model)
+        c = ControlNet(self.control_model, global_average_pooling=self.global_average_pooling)
         c.cond_hint_original = self.cond_hint_original
         c.strength = self.strength
         return c
@@ -790,7 +794,11 @@ def load_controlnet(ckpt_path, model=None):
     if use_fp16:
         control_model = control_model.half()
 
-    control = ControlNet(control_model)
+    global_average_pooling = False
+    if ckpt_path.endswith("_shuffle.pth") or ckpt_path.endswith("_shuffle.safetensors") or ckpt_path.endswith("_shuffle_fp16.safetensors"): #TODO: smarter way of enabling global_average_pooling
+        global_average_pooling = True
+
+    control = ControlNet(control_model, global_average_pooling=global_average_pooling)
     return control
 
 class T2IAdapter: