🚗 Exemples d'utilisation - Car Damage Detection

📸 Exemple 1 : Analyse simple

from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image

model = AutoModelForCausalLM.from_pretrained("Kakyoin03/car-damage-detection-llama-vision-14k")
tokenizer = AutoTokenizer.from_pretrained("Kakyoin03/car-damage-detection-llama-vision-14k")

image = Image.open("car_damage.jpg")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "text", "text": "Décrivez les dommages sur cette voiture."},
            {"type": "image", "image": image}
        ]
    }
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
outputs = model.generate(inputs, max_new_tokens=300)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

🔄 Exemple 2 : Traitement par batch

import torch
from PIL import Image
import glob

def analyze_multiple_cars(image_paths):
    results = []
    for path in image_paths:
        image = Image.open(path)
        # ... (même code que l'exemple 1)
        results.append({"image": path, "analysis": response})
    return results

# Analyser tous les JPG dans un dossier
car_images = glob.glob("damage_photos/*.jpg")
analyses = analyze_multiple_cars(car_images)

🎛️ Exemple 3 : Paramètres avancés

# Configuration avancée pour analyses détaillées
generation_config = {
    "max_new_tokens": 500,
    "temperature": 0.1,
    "top_p": 0.9,
    "do_sample": True,
    "repetition_penalty": 1.1
}

outputs = model.generate(inputs, **generation_config)