ex.py

from PIL import Image
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

# %%

model = Qwen2VLForConditionalGeneration.from_pretrained(pretrained_model_name_or_path="Qwen/Qwen2-VL-2B-Instruct",
                                                        torch_dtype="auto",
                                                        device_map="auto")

processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")


# %%
def rescale_image_dimensions(original_width, original_height, max_size=1000):
    # Orijinal boyutlar 1000 pikselin üzerindeyse yeniden ölçeklendir
    if original_width > max_size or original_height > max_size:
        aspect_ratio = original_width / original_height

        if aspect_ratio > 1:  # Genişlik yükseklikten büyükse
            scaled_width = max_size
            scaled_height = int(max_size / aspect_ratio)
        else:  # Yükseklik genişlikten büyükse veya eşitse
            scaled_height = max_size
            scaled_width = int(max_size * aspect_ratio)
    else:
        # Orijinal boyutlar zaten uygun ise
        scaled_width = original_width
        scaled_height = original_height

    return scaled_width, scaled_height


# %%
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "/home/nuh-hatipoglu/Desktop/NewMind/demo-vit/images/IMG_3644.JPG",
            },
            {"type": "text", "text": "Describe image"},
        ],
    }
]
# %%%
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# %%
image_inputs, video_inputs = process_vision_info(messages)

# %%

original_width, original_height = image_inputs[0].size
new_width, new_height = rescale_image_dimensions(original_width, original_height)
rescaled_image = image_inputs[0].resize((new_width, new_height), Image.Resampling.LANCZOS)
image_inputs = rescaled_image

#%%

image_inputs[0].show()
# %%
inputs = processor(text=[text],
                   images=image_inputs,
                   videos=video_inputs,
                   padding=True,
                   return_tensors="pt", )
inputs = inputs.to("cuda")
# %%
# Görseli aç
image_path = "your_image_path.jpg"  # Görselin dosya yolu
image = Image.open(image_path)

# Orijinal boyutları al
original_width, original_height = image.size

# Yeni boyutları hesapla
new_width, new_height = rescale_image_dimensions(original_width, original_height)

# Görseli yeniden boyutlandır
rescaled_image = image.resize((new_width, new_height), Image.ANTIALIAS)

# Yeniden boyutlandırılmış görseli kaydet
rescaled_image.save("rescaled_" + image_path)

print(f"Görsel başarıyla yeniden boyutlandırıldı: {new_width}x{new_height}")