from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

def load_llm():
    
    # default: Load the model on the available device(s)
    model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
        "Qwen/Qwen2.5-VL-3B-Instruct", torch_dtype="auto", device_map="auto"
    )
    min_pixels = 256*28*28
    max_pixels = 1280*28*28
    processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)
    return model,processor

import cv2
import base64
from PIL import Image
from io import BytesIO

def check_gender(model,processor,img_list,test_image):
    content = [
                {"type": "text", "text": "Both of the images belong to same person. Identify the gender and respond only Male or Female"},
            ]
    # Load image and crop face
    img = cv2.imread(test_image)
    for image in img_list[:3]:
        x1, y1, x2, y2 = image['bbox']
        face_crop = img[y1:y2, x1:x2]
        # Encode to JPEG format in memory
        _, buffer = cv2.imencode('.jpg', face_crop)
        # Convert to Base64
        face_base64 = base64.b64encode(buffer).decode('utf-8')
        image_bytes = base64.b64decode(face_base64)
        image_64 = Image.open(BytesIO(image_bytes)).convert("RGB")
        content.append({"type": "image", "image": image_64})
    messages = [
        {
            "role": "user",
            "content": content
        }
    ]

    # Preparation for inference
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    )
    inputs = inputs.to("cuda")

    # Inference
    generated_ids = model.generate(**inputs, max_new_tokens=128)
    generated_ids_trimmed = [
        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )
    print(output_text)
    return output_text

def visual_llm_gender(unique_chars,predictions,test_image):
    model,processor = load_llm()
    # for panel in predictions.faces:
    #     print(panel)
    gender_list = []
    for i in range(unique_chars):
        print(i)
        new_panels = [p for p in predictions.faces if p['char_id'] == i]
        print(new_panels)
        gender = check_gender(model,processor,new_panels,test_image)
        gender_list.append(gender)
    return gender_list