파이콘 한국 2025 심각한 중독입니다

python-yolo.py

from ultralytics import YOLO
from PIL import Image, ImageFont, ImageDraw
import numpy as np
import cv2
import textwrap

# 입력 이미지 및 모델
image_path = "img800.jpg"
model = YOLO("yolov8m-seg-speech-bubble.pt")
font_path = "ko.otf"
text_to_insert = [
    "흥, 웃기는 소리. 파이콘 한국 2025 pycon.kr 8월 15일부터 17일까지 동국대학교 서울캠퍼스 신공학관에서",
    "치료가 필요할 정도로 심각한 파이썬 중독입니다",
]

# 이미지 불러오기
image = Image.open(image_path).convert("RGB")
image_np = np.array(image)
h, w = image_np.shape[:2]

def fits_in_contour(lines, font, contour, origin):
    draw = ImageDraw.Draw(Image.new("RGB", (1,1)))
    bbox = draw.multiline_textbbox((0, 0), "\n".join(lines), font=font)
    text_w, text_h = bbox[2] - bbox[0], bbox[3] - bbox[1]
    tx, ty = origin[0] - text_w // 2, origin[1] - text_h // 2

    for dy in range(0, text_h, 2):
        for dx in range(0, text_w, 2):
            px, py = float(tx + dx), float(ty + dy)
            if cv2.pointPolygonTest(contour, (px, py), False) < 0:
                return False
    return True

def find_best_wrapped_text(text, font_path, contour, max_font_size):
    M = cv2.moments(contour)
    if M["m00"] == 0:
        return None
    center = (int(M["m10"] / M["m00"]), int(M["m01"] / M["m00"]))
    best_result = None

    for font_size in range(max_font_size, 10, -1):
        font = ImageFont.truetype(font_path, font_size)
        for width in range(5, len(text) + 1):  # 최대 너비
            wrapped = textwrap.fill(text, width=width)
            lines = wrapped.splitlines()
            if fits_in_contour(lines, font, contour, center):
                best_result = (lines, font, center)
                break
        if best_result:
            break

    return best_result  # (lines, font, center)

# 예측 실행
results = model.predict(source=image_path, save=False, imgsz=max(h, w))

for r in results:
    if r.masks is None:
        continue

    class_names = model.names
    target_class_name = "speech bubble"

    # 마스크, 클래스 ID, confidence 모두 순회
    for i, (mask, cls_id, conf, box) in enumerate(zip(r.masks.data, r.boxes.cls, r.boxes.conf, r.boxes.xyxy)):
        class_id = int(cls_id.item())
        if class_names[class_id] != target_class_name:
            continue

        # 박스
        x1, y1, x2, y2 = map(int, box.tolist())
        cv2.rectangle(image_np, (x1, y1), (x2, y2), (0, 255, 0), 2)
        label = f"{class_names[class_id]} {conf.item():.2f}"
        cv2.putText(image_np, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 128, 255), 1)

        # 마스크 윤곽선 추출
        mask_np = mask.cpu().numpy()
        mask_resized = cv2.resize(mask_np, (w, h), interpolation=cv2.INTER_NEAREST)
        binary = (mask_resized > 0.5).astype(np.uint8)
        contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

        for contour in contours:
            M = cv2.moments(contour)
            if M["m00"] == 0:
                continue
            cx = int(M["m10"] / M["m00"])
            cy = int(M["m01"] / M["m00"])

            contour = contour.astype(np.float32)
            contour -= [cx, cy]
            contour *= 0.95
            contour += [cx, cy]
            contour = contour.astype(np.int32)

            # 하얀색으로 채움 (지우기)
            cv2.drawContours(image_np, [contour], -1, (255, 255, 255), thickness=cv2.FILLED)

            # 텍스트 그릴 공간 계산
            result = find_best_wrapped_text(text_to_insert[i], font_path, contour, 40)
            if result:
                lines, font, center = result
                pil_img = Image.fromarray(image_np)
                draw = ImageDraw.Draw(pil_img)
                draw.multiline_text(center, "\n".join(lines), font=font, fill=(0, 0, 0), anchor="mm", align="center")
                image_np = np.array(pil_img)

# 저장 및 표시
Image.fromarray(image_np).show()

model: https://huggingface.co/kitsumed/yolov8m_seg-speech-bubble