.history/mbench_a2d/gpt_a2d_numbered_20250205111640.py

from datasets import build_dataset
import argparse
import opts

import sys
import os
import time

import numpy as np
import matplotlib.pyplot as plt
import cv2
from io import BytesIO
import base64
from PIL import Image

from openai import OpenAI

def mark_object_and_encode(frame, mask, instance_id, text_query, color_mask=False, label_number=False):
    #마스크 색칠할지
    if color_mask == True:
        alpha = 0.1

        colored_mask = np.zeros_like(frame)
        colored_mask[mask == 1] = [255, 0, 0]
        frame[mask == 1] = (
            (1 - alpha) * frame[mask == 1] +
            alpha * colored_mask[mask == 1]
        )

    #마스크 아웃라인 그리기
    contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    cv2.drawContours(frame, contours, -1, [255, 0, 0], 2)

    #instance_id 적을지
    if label_number == True:
        if len(contours) > 0:
            largest_contour = max(contours, key=cv2.contourArea)
            M = cv2.moments(largest_contour)
            if M["m00"] != 0:  
                center_x = int(M["m10"] / M["m00"])
                center_y = int(M["m01"] / M["m00"])
            else:
                center_x, center_y = 0, 0

            font = cv2.FONT_HERSHEY_SIMPLEX
            text = str(instance_id)
            font_scale = 0.6  
            text_size = cv2.getTextSize(text, font, font_scale, 2)[0]                        
            text_x = center_x - text_size[0] // 1  # 텍스트의 가로 중심
            text_y = center_y
            # text_y = center_y + text_size[1] // 2 # 텍스트의 세로 중심
            
            # 텍스트 배경 사각형 좌표 계산
            rect_start = (text_x - 5, text_y - text_size[1] - 5) # 배경 사각형 좌상단
            # rect_end = (text_x + text_size[0] + 5, text_y + 5) 
            rect_end = (text_x + text_size[0] + 5, text_y)
            
            cv2.rectangle(frame, rect_start, rect_end, (0, 0, 0), -1)
            cv2.putText(frame, text, (text_x, text_y), font, font_scale, (255, 255, 255), 2)

    # plt.figure(figsize=(6, 10))
    # plt.imshow(frame)
    # plt.title(text_query)
    # plt.tight_layout()
    # plt.axis('off')
    # plt.show()

    buffer = BytesIO()
    frame = Image.fromarray(frame)
    frame.save(buffer, format='jpeg')
    buffer.seek(0)
    encoded_frame = base64.b64encode(buffer.read()).decode("utf-8")

    return encoded_frame


if __name__ == "__main__":
    parser = argparse.ArgumentParser('ReferFormer training and evaluation script', parents=[opts.get_args_parser()])
    args = parser.parse_args()

    train_dataset = build_dataset('a2d', image_set = 'train', args = args)
    text_annotations = train_dataset.text_annotations