app.py

import torch
import torch.nn as nn
import torchvision.models as models
import torchvision.transforms as transforms
from PIL import Image
import gradio as gr

# --------------------
# Class Mapping
# --------------------
class_to_idx = {
    'Acura': 0, 'Alfa Romeo': 1, 'Aston Martin': 2, 'Audi': 3, 'BMW': 4,
    'Bentley': 5, 'Bugatti': 6, 'Buick': 7, 'Cadillac': 8, 'Chevrolet': 9,
    'Chrysler': 10, 'Citroen': 11, 'Daewoo': 12, 'Dodge': 13, 'Ferrari': 14,
    'Fiat': 15, 'Ford': 16, 'GMC': 17, 'Genesis': 18, 'Honda': 19,
    'Hudson': 20, 'Hyundai': 21, 'Infiniti': 22, 'Jaguar': 23, 'Jeep': 24,
    'Kia': 25, 'Land Rover': 26, 'Lexus': 27, 'Lincoln': 28, 'MG': 29,
    'Maserati': 30, 'Mazda': 31, 'Mercedes-Benz': 32, 'Mini': 33,
    'Mitsubishi': 34, 'Nissan': 35, 'Oldsmobile': 36, 'Peugeot': 37,
    'Pontiac': 38, 'Porsche': 39, 'Ram Trucks': 40, 'Renault': 41,
    'Saab': 42, 'Studebaker': 43, 'Subaru': 44, 'Suzuki': 45, 'Tesla': 46,
    'Toyota': 47, 'Volkswagen': 48, 'Volvo': 49
}
idx_to_class = {v: k for k, v in class_to_idx.items()}

# --------------------
# Image Transform
# --------------------
transform = transforms.Compose([
    transforms.Lambda(lambda x: x.convert("RGB")),
    transforms.Resize((224,224)),
    transforms.RandomHorizontalFlip(p=0.5),
    transforms.RandomVerticalFlip(p=0.2),
    transforms.RandomRotation(20),
    transforms.ColorJitter(brightness=0.3, contrast=0.3, saturation=0.3),
    transforms.RandomResizedCrop(224, scale=(0.7, 1.0)),
    transforms.ToTensor(),
    transforms.Normalize([0.5]*3, [0.5]*3)
])

# --------------------
# Load Model
# --------------------
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load pretrained ResNet50 correctly
base_model = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)

# Replace final fully connected layer (your head)
in_features = base_model.fc.in_features  # 2048 for ResNet50
base_model.fc = nn.Sequential(
    nn.Linear(in_features, 512),
    nn.ReLU(),
    nn.Dropout(0.5),
    nn.Linear(512, 50)  # 50 classes
)

# Load state dict
state_dict = torch.load("best_model.pth", map_location=device)
base_model.load_state_dict(state_dict)
base_model = base_model.to(device)
base_model.eval()

# --------------------
# Prediction Function
# --------------------
def predict(img):
    img = transform(img).unsqueeze(0).to(device)

    with torch.no_grad():
        outputs = base_model(img)
        probs = torch.softmax(outputs, dim=1)[0]

    top5_prob, top5_idx = torch.topk(probs, 5)
    result = {idx_to_class[idx.item()]: float(top5_prob[i]) for i, idx in enumerate(top5_idx)}
    return result

# --------------------
# Gradio UI
# --------------------
demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=5),
    title="Car Brand Classifier",
    description="Upload a car image to predict its brand."
)

if __name__ == "__main__":
    demo.launch()