Upload app.py

app.py

@@ -0,0 +1,51 @@
+# app.py
+import gradio as gr
+import os
+
+# Import the core logic from your "normal inference file"
+from NoiseFilter.MaIR.inference_runner import MaIR_Upsampler
+
+# --- Global model cache for performance ---
+# This dictionary will store loaded models to avoid reloading on every API call.
+model_cache = {}
+
+def get_model(model_name):
+    """Loads a model into the cache if it's not already there."""
+    if model_name not in model_cache:
+        print(f"Loading model {model_name} into cache...")
+        model_cache[model_name] = MaIR_Upsampler(model_name=model_name)
+    return model_cache[model_name]
+
+# --- API Function ---
+def inference_api(image, model_name):
+    """
+    This is the function that the API will call.
+    It takes a NumPy array and a model name string as input.
+    """
+    if image is None:
+        # Gradio handles this by not running, but good practice for raw API calls.
+        raise ValueError("No image provided.")
+    
+    upsampler = get_model(model_name)
+    output_image = upsampler.process(image)
+    return output_image
+
+# --- Create the Gradio Interface (for API generation) ---
+# We define a minimal interface. The primary goal is API exposure.
+interface = gr.Interface(
+    fn=inference_api,
+    inputs=[
+        gr.Image(type="numpy", label="Input Image"),
+        gr.Dropdown(
+            choices=['MaIR-SRx4', 'MaIR-SRx2', 'MaIR-CDN-s50'], 
+            value='MaIR-SRx4', 
+            label="Select Model"
+        ),
+    ],
+    outputs=gr.Image(type="numpy", label="Output Image"),
+    title="MaIR: Image Restoration API",
+    description="API for MaIR models. Use the '/api' endpoint for programmatic access."
+)
+
+# Launch the app. This will start the web server and create the API.
+interface.launch()