txt

app.py

@@ -200,10 +200,20 @@ generate_random_imgs_from_dataset_partial = partial(generate_random_imgs_from_da
 print(f"[Init] CUDA max allocated: {torch.cuda.max_memory_allocated() / 1024**2:.2f} MB")
 print(f"[Init] CUDA max reserved: {torch.cuda.max_memory_reserved() / 1024**2:.2f} MB")
 
-title = "Inverse problem playground"  # displayed on gradio tab and in the gradio page
+title = "Reconstruct Anything Model Demo"  # displayed on gradio tab and in the gradio page
 with gr.Blocks(title=title, theme=gr.themes.Glass()) as interface:
     gr.Markdown("## " + title)
 
+    gr.Markdown(
+        """
+        This demo showcases the performance of the **Reconstruct Anything Model (RAM)** across a variety of inverse problems on both natural and MRI images.  
+        Select a dataset and a physics task below (e.g., inpainting, super-resolution, deblurring...).  
+        Note: The parameters of the selected physics — such as noise levels, blur kernels, or inpainting masks — are randomly generated before reconstruction, leveraging the [deepinverse library](https://deepinv.github.io/deepinv/).  
+
+        📄 For more details on the method, check out our [paper on arXiv](https://arxiv.org/abs/2503.08915).
+        """
+    )
+
     ### USER-SPECIFIC VARIABLES
     dataset_placeholder = gr.State(get_dataset_on_DEVICE_STR("Natural"))
     available_physics_placeholder = gr.State(['Inpainting', 'SR', 'MotionBlur_medium', 'MotionBlur_hard',