Create app.py

app.py

@@ -0,0 +1,214 @@
+import gradio as gr
+from PIL import Image
+import os
+import tempfile
+import numpy as np
+import traceback
+from tensorflow.keras.models import load_model
+from tensorflow.keras import backend as K
+from tensorflow.keras.preprocessing import image as keras_image
+
+print("Initializing Gradio app...")
+
+# Update this path to match where your .h5 models actually live
+MODEL_DIR = r"D:\ML_Bench\License_Plate\mlp\vehicle-rear-master\models"
+
+
+def get_model_choices():
+    if not os.path.exists(MODEL_DIR):
+        print(f"[ERROR] Model directory not found: {MODEL_DIR}")
+        return [], MODEL_DIR
+
+    model_files = [f for f in os.listdir(MODEL_DIR) if f.lower().endswith('.h5')]
+    model_files.sort()
+    print(f"[INFO] Found model files: {model_files}")
+    return model_files, MODEL_DIR
+
+
+def process_load(img_path, shape_tuple):
+    """
+    Load and convert the image at img_path to a numpy array
+    of shape (height, width, 3) matching shape_tuple, scaled to [0,1].
+    """
+    try:
+        target_h, target_w = shape_tuple[0], shape_tuple[1]
+        print(f"[DEBUG] process_load: loading '{img_path}' at size ({target_h}, {target_w})")
+        img = keras_image.load_img(img_path, target_size=(target_h, target_w))
+        x = keras_image.img_to_array(img)
+        print(f"[DEBUG] process_load: result array shape {x.shape}")
+        return x
+    except Exception as e:
+        print(f"[ERROR] process_load failed for {img_path}: {e}")
+        traceback.print_exc()
+        raise
+
+
+def predict(img1, img2, model_file_name):
+    print("\n[DEBUG] predict() called.")
+    print(f"        img1 type: {type(img1)}, img2 type: {type(img2)}, model_file_name: {model_file_name}")
+
+    # 1. Make sure both images are uploaded
+    if img1 is None or img2 is None:
+        print("[ERROR] One or both images were not uploaded.")
+        return "Error: You must upload both Image 1 and Image 2."
+
+    # 2. Get list of available models each time
+    model_files, model_dir = get_model_choices()
+    if not model_files:
+        print("[ERROR] No .h5 model files found in directory.")
+        return f"Error: No .h5 model files found in {MODEL_DIR}."
+
+    # 3. Make sure a model was actually selected
+    if model_file_name is None or model_file_name not in model_files:
+        print(f"[ERROR] Invalid or missing model selection: {model_file_name}")
+        return "Error: Please select a valid model from the dropdown."
+
+    model_path = os.path.join(model_dir, model_file_name)
+    print(f"[INFO] Loading model from: {model_path}")
+
+    img1_path = img2_path = None
+    try:
+        # 4. Infer input size based on filename convention (defensively)
+        basename = os.path.splitext(model_file_name)[0]
+        parts = basename.split('_')
+        name = ""
+        if len(parts) >= 3 and parts[0].lower() == "model" and parts[1].lower() == "shape":
+            name = parts[2].lower()
+            print(f"[DEBUG] Inferred model base name: {name}")
+        else:
+            name = "smallvgg"
+            print(f"[DEBUG] Filename did not match 'model_shape_X'; using default '{name}'")
+
+        if name in ['resnet50', 'vgg16']:
+            image_size_h_c, image_size_w_c = 224, 224
+        elif name == 'googlenet':
+            image_size_h_c, image_size_w_c = 112, 112
+        else:
+            image_size_h_c, image_size_w_c = 128, 128
+
+        input_shape = (image_size_h_c, image_size_w_c, 3)
+        print(f"[INFO] Inferred input_shape = {input_shape} based on '{name}'")
+
+        # 5. Save PIL images to temporary files
+        try:
+            with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as f1:
+                img1_path = f1.name
+                img1.save(img1_path)
+            print(f"[DEBUG] img1 saved to: {img1_path} ({os.path.getsize(img1_path)} bytes)")
+
+            with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as f2:
+                img2_path = f2.name
+                img2.save(img2_path)
+            print(f"[DEBUG] img2 saved to: {img2_path} ({os.path.getsize(img2_path)} bytes)")
+
+        except Exception as e:
+            print(f"[ERROR] Failed to save uploaded images: {e}")
+            traceback.print_exc()
+            raise
+
+        # 6. Preprocess images
+        try:
+            img1_arr = process_load(img1_path, input_shape) / 255.0
+            img2_arr = process_load(img2_path, input_shape) / 255.0
+            img1_batch = np.expand_dims(img1_arr, axis=0)
+            img2_batch = np.expand_dims(img2_arr, axis=0)
+            print(f"[DEBUG] Prepared batches: img1_batch.shape = {img1_batch.shape}, img2_batch.shape = {img2_batch.shape}")
+        except Exception as e:
+            print(f"[ERROR] Preprocessing failed: {e}")
+            traceback.print_exc()
+            raise
+
+        # 7. Clear previous Keras session (avoid GPU OOM if repeated calls)
+        print("[DEBUG] Clearing Keras session.")
+        K.clear_session()
+
+        # 8. Load model
+        try:
+            print("[DEBUG] Calling load_model...")
+            model = load_model(model_path, compile=False)
+            print(f"[INFO] Model loaded successfully. Model inputs: {model.input_shape}")
+        except Exception as e:
+            error_str = f"Prediction failed: could not load model '{model_file_name}': {e}"
+            print(f"[ERROR] {error_str}")
+            traceback.print_exc()
+            return error_str
+
+        # 9. Perform prediction
+        try:
+            print("[DEBUG] Calling model.predict...")
+            Y_ = model.predict([img1_batch, img2_batch])
+            print(f"[DEBUG] Raw model output: {Y_}")
+        except Exception as e:
+            error_str = f"Prediction failed during model.predict: {e}"
+            print(f"[ERROR] {error_str}")
+            traceback.print_exc()
+            return error_str
+
+        # 10. Interpret prediction
+        try:
+            pred_idx = int(np.argmax(Y_[0]))
+            pred = 'positive' if pred_idx == 1 else 'negative'
+            confidence = float(np.max(Y_[0]))
+            result_str = f"Prediction: {pred} (confidence: {confidence:.2f})"
+            print(f"[INFO] {result_str}")
+            return result_str
+        except Exception as e:
+            error_str = f"Prediction failed while interpreting output: {e}"
+            print(f"[ERROR] {error_str}")
+            traceback.print_exc()
+            return error_str
+
+    except Exception as e:
+        # Catches any “unexpected” error we didn’t already return
+        error_msg = f"Prediction failed with unexpected error: {str(e)}"
+        print(f"[ERROR] {error_msg}")
+        traceback.print_exc()
+        return error_msg
+
+    finally:
+        # 11. Cleanup temp files
+        for path in [img1_path, img2_path]:
+            if path and os.path.exists(path):
+                try:
+                    os.remove(path)
+                    print(f"[DEBUG] Deleted temp file: {path}")
+                except Exception as e:
+                    print(f"[WARNING] Could not delete temp file {path}: {e}")
+                    traceback.print_exc()
+
+
+def main():
+    model_files, _ = get_model_choices()
+    if not model_files:
+        print(f"[ERROR] No model files found. Please put .h5 files in {MODEL_DIR} before running.")
+    else:
+        print(f"[INFO] Models available: {model_files}")
+
+    with gr.Blocks() as demo:
+        gr.Markdown("# Siamese Shape Stream Prediction App")
+        gr.Markdown("Upload two images and select a model to get a similarity prediction.")
+
+        with gr.Row():
+            img1 = gr.Image(label="Image 1", type="pil")
+            img2 = gr.Image(label="Image 2", type="pil")
+
+        model_dropdown = gr.Dropdown(
+            choices=model_files,
+            value=model_files[0] if model_files else None,
+            label="Model File (.h5)",
+            interactive=True
+        )
+
+        predict_btn = gr.Button("Predict")
+        output = gr.Textbox(label="Prediction Output", lines=2)
+
+        predict_btn.click(
+            fn=predict,
+            inputs=[img1, img2, model_dropdown],
+            outputs=output
+        )
+
+    demo.launch(share=True, server_name="0.0.0.0", server_port=7860, debug=True, enable_queue=True)
+
+if __name__ == "__main__":
+    main()