app.py

import torch
import gradio as gr
from transformers import ViTImageProcessor, AutoModelForImageClassification
from PIL import Image
import numpy as np
import time

# -----------------------------
# Configuration and Setup
# -----------------------------

# Force Gradio to use CUDA (if available)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# Model path
model_path = "final_model"

# Load image processor and model
try:
    print("Loading image processor...")
    processor = ViTImageProcessor.from_pretrained(model_path)

    print("Loading model...")
    model = AutoModelForImageClassification.from_pretrained(model_path)
    model = model.to(device)
    model.eval()  # Important for deterministic behavior
except Exception as e:
    raise RuntimeError(f"Error loading model: {e}")

# Attempt to load label mappings
try:
    labels = model.config.id2label
    assert isinstance(labels, dict) and len(labels) > 0, "Invalid or empty id2label mapping"
except Exception as e:
    print(f"⚠️ Labels not found in model config: {e}")
    labels = {i: f"Class {i}" for i in range(model.config.num_labels)}


# -----------------------------
# Standalone Test Mode (Optional)
# -----------------------------
def test_inference():
    """Run inference outside Gradio to verify model works"""
    dummy_img = Image.new('RGB', (224, 224), color='red')  # Create a dummy image
    print("Running standalone inference test...")
    try:
        inputs = processor(images=dummy_img, return_tensors="pt").to(device)
        with torch.inference_mode():
            outputs = model(**inputs)
        print("✅ Model inference test successful")
    except Exception as e:
        print(f"❌ Inference test failed: {e}")


# -----------------------------
# Prediction Function
# -----------------------------

def predict(image):
    if image is None:
        return "No image uploaded."

    print("\n[INFO] Starting prediction pipeline...")

    # Step 1: Preprocessing
    print("[STEP 1] Preprocessing image...")
    try:
        start = time.time()
        inputs = processor(images=image, return_tensors="pt").to(device)
        print(f"[DEBUG] Input shape: {inputs['pixel_values'].shape}")
        print(f"[DEBUG] Time taken: {time.time() - start:.2f}s")
    except Exception as e:
        return f"❌ Error in preprocessing: {e}"

    # Step 2: Inference
    print("[STEP 2] Running inference...")
    try:
        start = time.time()
        with torch.inference_mode():
            outputs = model(**inputs)
        print(f"[DEBUG] Inference completed in {time.time() - start:.2f}s")
    except Exception as e:
        return f"❌ Error in model inference: {e}"

    # Step 3: Post-processing
    print("[STEP 3] Processing output...")
    try:
        probs = torch.nn.functional.softmax(outputs.logits, dim=1)
        top5_probs, top5_indices = torch.topk(probs, 5)

        result = ""
        for i in range(5):
            idx = top5_indices[0][i].item()
            label = labels.get(idx, f"Unknown class {idx}")
            prob = top5_probs[0][i].item() * 100
            result += f"{i + 1}. {label} — {prob:.2f}%\n"
    except Exception as e:
        return f"❌ Error post-processing: {e}"

    print("[INFO] Prediction complete ✅\n")
    return result.strip()


# -----------------------------
# Gradio Interface
# -----------------------------

interface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil", label="Upload an Image"),
    outputs=gr.Textbox(label="Top 5 Predictions"),
    title="Fine-Tuned ViT Image Classifier",
    description="Upload an image to get the top 5 predicted classes with confidence scores.",
    allow_flagging="never",
    examples=[["examples/test_image.jpg"]] if "examples" in locals() else None
)

if __name__ == "__main__":
    print("\n🚀 Launching Gradio interface...\n")
    test_inference()  # Optional: Run test before launching
    interface.launch(share=True)