import gradio as gr
import os
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision.models as models
from torchvision.models import ResNet18_Weights
from PIL import Image
from timeit import default_timer as timer
from typing import Tuple, Dict

# 1. Class names (order must match your training)
class_names = ['Cabbage', 'Carrot', 'Cucumber', 'Tomato']

# 2. Load default preprocessing from pretrained weights
weights = ResNet18_Weights.DEFAULT
transform = weights.transforms()

# 3. Define and load the model
def get_resnet18():
    model = models.resnet18(weights=weights)
    num_ftrs = model.fc.in_features
    model.fc = nn.Linear(num_ftrs, len(class_names))
    return model

model = get_resnet18()
model.load_state_dict(torch.load("resnet_model.pth", map_location=torch.device("cpu"), weights_only=False))
model.eval()

# 4. Prediction function
def predict(img: Image.Image) -> Tuple[Dict[str, float], float]:
    start = timer()
    img_tensor = transform(img).unsqueeze(0)

    with torch.inference_mode():
        output = model(img_tensor)
        probs = torch.softmax(output, dim=1)[0]

    result = {class_names[i]: float(probs[i]) for i in range(len(class_names))}
    elapsed = round(timer() - start, 5)
    return result, elapsed

# 5. Build the Gradio interface
with gr.Blocks() as demo:
    gr.HTML("""
        <div style="background-image: url('https://images.unsplash.com/photo-1501004318641-b39e6451bec6');
                    background-size: cover; 
                    padding: 25px;
                    border-radius: 10px;
                    color: white;
                    font-family: 'Poppins', sans-serif;">
            <h1 style='text-align: center;'>🌾 Vegetable Classifier with ResNet18 🌾</h1>
            <h3 style='text-align: center;'>
                Created by <a href='https://github.com/ExplorerGumel' target='_blank' style='color: #f0e68c;'>@ExplorerGumel</a>
            </h3>
        </div>
    """)

    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="📸 Upload Vegetable Image")
            predict_button = gr.Button("🔍 Predict")

            
            examples = []
            if os.path.exists("examples"):
                examples = [["examples/" + fname] for fname in os.listdir("examples") if fname.lower().endswith((".png", ".jpg", ".jpeg"))]
            gr.Examples(examples=examples, inputs=[image_input])

        with gr.Column():
            output_label = gr.Label(num_top_classes=4, label="🌽 Predictions")
            output_time = gr.Number(label="⏱️ Prediction Time (seconds)")

    predict_button.click(fn=predict, inputs=image_input, outputs=[output_label, output_time])

# 6. Launch the app
demo.launch()