from transformers import BlipProcessor, BlipForQuestionAnswering import torch from PIL import Image device = "cuda" if torch.cuda.is_available() else "cpu" # Chargement du modèle BLIP pré-entraîné (VQA) processor = BlipProcessor.from_pretrained("Salesforce/blip-vqa-base", use_fast=True) model = BlipForQuestionAnswering.from_pretrained("Salesforce/blip-vqa-base").to(device) model.eval() def ask_blip(image_path, question): image = Image.open(image_path).convert("RGB") # Préparer les inputs (image + question) inputs = processor(image, question, return_tensors="pt").to(device) # Générer la réponse with torch.no_grad(): outputs = model.generate(**inputs) answer = processor.tokenizer.decode(outputs[0], skip_special_tokens=True) return answer