test

app.py

@@ -312,7 +312,7 @@ def create_demo(model):
 
         def generate_image(*inputs):
             images = []
-            if grid_h != model.grid_h or grid_w != model.grid_w:
+            if grid_h + 1 != model.grid_h or grid_w != model.grid_w:
                 raise gr.Error('Please wait for the loading to complete.')
             for i in range(model.grid_h):
                 images.append([])

demo_tasks/gradio_tasks.py

@@ -131,7 +131,7 @@ def process_dense_prediction_tasks(x):
             mask = task.get('mask', [0 for _ in range(grid_w - 1)] + [1])
             layout_prompt = get_layout_instruction(grid_w, grid_h)
             
-            upsampling_noise = 0.7
+            upsampling_noise = 1.0
             steps = None
             outputs = [mask, grid_h, grid_w, layout_prompt, task_prompt, content_prompt, upsampling_noise, steps] + rets
             break

demo_tasks/gradio_tasks_unseen.py

@@ -253,7 +253,7 @@ def process_unseen_tasks(x):
             mask = task.get('mask', [0 for _ in range(grid_w - 1)] + [1])
             layout_prompt = get_layout_instruction(grid_w, grid_h)
             
-            upsampling_noise = 0.7
+            upsampling_noise = 1.0
             steps = None
             outputs = [mask, grid_h, grid_w, layout_prompt, task_prompt, content_prompt, upsampling_noise, steps] + rets
             break

visualcloze.py

@@ -190,6 +190,8 @@ class VisualClozeModel:
         )
         
         image = image.resize(((target_size[0] // 16) * 16, (target_size[1] // 16) * 16))
+        if upsampling_noise >= 1.0:
+            return image
         processed_image = self.image_transform(image)
         processed_image = processed_image.to(self.device, non_blocking=True)
         blank = torch.zeros_like(processed_image, device=self.device, dtype=self.dtype)