add resize img & convert to complex and to grayscale when necessary

app.py

@@ -21,6 +21,16 @@ DEVICE_STR = 'cuda' if torch.cuda.is_available() else 'cpu'
 
 
 ### Gradio Utils
+def resize_tensor_within_box(tensor_img: torch.Tensor, max_size: int = 512):
+    _, _, h, w = tensor_img.shape
+    scale = min(max_size / h, max_size / w)
+
+    if scale < 1.0:
+        new_h, new_w = int(h * scale), int(w * scale)
+        tensor_img = transforms.functional.resize(tensor_img, [new_h, new_w], antialias=True)
+
+    return tensor_img
+
 def generate_imgs_from_user(image,
                             physics: PhysicsWithGenerator, use_gen: bool,
                             baseline: BaselineModel, model: EvalModel,
@@ -31,9 +41,17 @@ def generate_imgs_from_user(image,
 
     # PIL image -> torch.Tensor / (1, C, H, W) / move to DEVICE_STR
     x = transforms.ToTensor()(image).unsqueeze(0).to(DEVICE_STR)
+    # Resize img within a 512x512 box
+    x = resize_tensor_within_box(x)
 
-    return generate_imgs(x, physics, use_gen, baseline, model, metrics)
+    C = x.shape[1]
+    if C == 3 and physics.name == 'CT':
+        x = transforms.Grayscale(num_output_channels=1)(x)
+    elif C == 3 and physics.name == 'MRI':  # not working because MRI physics has a fixed img size
+        x = transforms.Grayscale(num_output_channels=1)(x)
+        x = torch.cat((x, torch.zeros_like(x)), dim=1)
 
+    return generate_imgs(x, physics, use_gen, baseline, model, metrics)
 def generate_imgs_from_dataset(dataset: EvalDataset, idx: int,
                                physics: PhysicsWithGenerator, use_gen: bool,
                                baseline: BaselineModel, model: EvalModel,
@@ -189,7 +207,8 @@ with gr.Blocks(title=title, theme=gr.themes.Glass()) as interface:
     print(f"[Render] CUDA max allocated: {torch.cuda.max_memory_allocated() / 1024**2:.2f} MB")
     print(f"[Render] CUDA max reserved: {torch.cuda.max_memory_reserved() / 1024**2:.2f} MB")
 
-    @gr.render(inputs=[dataset_placeholder, physics_placeholder, available_physics_placeholder])
+    @gr.render(inputs=[dataset_placeholder, physics_placeholder, available_physics_placeholder],
+               triggers=[dataset_placeholder.change, physics_placeholder.change])
     def dynamic_layout(dataset, physics, available_physics):
         ### LAYOUT