app.py

import gradio as gr
import torch
import torch.nn as nn
from torchvision import models
from torchvision.models import ResNet34_Weights
from PIL import Image
import torchvision.transforms as transforms
from huggingface_hub import hf_hub_download
import os
import random
import glob

# Import LoRA code
from model import LoRALayer, apply_lora_to_model

# Load model
print("Loading model...")
model = models.resnet34(weights=ResNet34_Weights.IMAGENET1K_V1)
model.fc = nn.Linear(model.fc.in_features, 2)
model = apply_lora_to_model(model, rank=8)

# Load trained weights (from local Space files)
model.load_state_dict(torch.load('best_model.pth', map_location='cpu'))
model.eval()
print("Model loaded successfully!")

# Preprocessing
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

# Class names
class_names = ['Non-Smoker', 'Smoker']

def predict(image):
    """
    Predict if person in image is smoking
    
    Args:
        image: PIL Image
        
    Returns:
        dict: Prediction probabilities for each class
    """
    if image is None:
        return None
    
    # Preprocess
    img_tensor = transform(image).unsqueeze(0)
    
    # Predict
    with torch.no_grad():
        outputs = model(img_tensor)
        probabilities = torch.softmax(outputs, dim=1)[0]
    
    # Format results
    results = {
        class_names[i]: float(probabilities[i]) 
        for i in range(len(class_names))
    }
    
    return results


# Get all example images
example_images = glob.glob("All/*")

examples = [[img] for img in example_images[:12]]  # Takes the 12 images


# Function to get random sample
def get_random_sample():
    """Load a random example image"""
    random_image_path = random.choice(example_images)
    return Image.open(random_image_path)
    
# Create Gradio interface with custom CSS
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🚬 Smoker Detection
        
        Upload an image or try a random sample to detect if a person is smoking.
        
        This model uses **ResNet34 with LoRA fine-tuning** (only 2.14% of parameters trained) 
        and achieves **89.73% test accuracy**.
        
        **Model:** [notrito/smoker-detection](https://huggingface.co/notrito/smoker-detection)
        """
    )
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Upload Image")
            
            with gr.Row():
                predict_btn = gr.Button("🔍 Predict", variant="primary")
                random_btn = gr.Button("🎲 Random Sample", variant="secondary")
            
        with gr.Column():
            output_label = gr.Label(num_top_classes=2, label="Prediction")
    
            gr.Markdown("### 📸 Try these examples:")
            gr.Examples(
                examples=examples,
                inputs=input_image,
                outputs=output_label,
                fn=predict,
                cache_examples=True
            )
    
    gr.Markdown(
        """
        ===================================================================================================
        
        ### About this model
        
        - **Architecture:** ResNet34 + LoRA adapters (rank=8)
        - **Training:** Fine-tuned on 1,120 images
        - **Performance:** 89.73% test accuracy, 89.96% F1-score
        - **Efficiency:** Only 465K trainable parameters (2.14% of model)
        
        ### How it works
        
        LoRA (Low-Rank Adaptation) freezes the pretrained ImageNet weights and adds small trainable 
        matrices to specific layers. This prevents overfitting on small datasets while maintaining 
        the model's powerful feature extraction capabilities.
        
        ### Limitations
        
        - Trained on limited dataset (1,120 images)
        - Best for frontal/profile views with visible cigarettes
        - May not generalize to all smoking scenarios
        
        ### Links
        
        - [Model Card](https://huggingface.co/notrito/smoker-detection)
        - [Training Notebook](https://www.kaggle.com/code/notrito/smoker-detection-with-lora)
        
        **Author:** Noel Triguero
        """
    )
    
    # Connect buttons
    predict_btn.click(fn=predict, inputs=input_image, outputs=output_label)
    random_btn.click(fn=get_random_sample, inputs=None, outputs=input_image)

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