| from PIL import Image | |
| from transformers import AutoProcessor, AutoModelForCausalLM | |
| def generate_caption(image_path, device="cuda"): | |
| print("Генерация подписи...") | |
| processor = AutoProcessor.from_pretrained("microsoft/git-base", use_fast=False) | |
| model = AutoModelForCausalLM.from_pretrained("microsoft/git-base").to(device) | |
| image = Image.open(image_path).convert("RGB") | |
| inputs = processor(images=image, return_tensors="pt").to(device) | |
| generated_ids = model.generate( | |
| pixel_values=inputs.pixel_values, | |
| max_length=50, | |
| pad_token_id=processor.tokenizer.pad_token_id | |
| ) | |
| caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0] | |
| print("Сгенерированная подпись:", caption) | |
| return caption |