| |
| |
|
|
| |
|
|
| |
| |
| |
|
|
| import argparse |
| import math |
| import cv2 |
| import glob |
| import os |
| from anime_face_detector import create_detector |
| from tqdm import tqdm |
| import numpy as np |
| from library.utils import setup_logging, pil_resize |
| setup_logging() |
| import logging |
| logger = logging.getLogger(__name__) |
|
|
| KP_REYE = 11 |
| KP_LEYE = 19 |
|
|
| SCORE_THRES = 0.90 |
|
|
|
|
| def detect_faces(detector, image, min_size): |
| preds = detector(image) |
| |
|
|
| faces = [] |
| for pred in preds: |
| bb = pred['bbox'] |
| score = bb[-1] |
| if score < SCORE_THRES: |
| continue |
|
|
| left, top, right, bottom = bb[:4] |
| cx = int((left + right) / 2) |
| cy = int((top + bottom) / 2) |
| fw = int(right - left) |
| fh = int(bottom - top) |
|
|
| lex, ley = pred['keypoints'][KP_LEYE, 0:2] |
| rex, rey = pred['keypoints'][KP_REYE, 0:2] |
| angle = math.atan2(ley - rey, lex - rex) |
| angle = angle / math.pi * 180 |
|
|
| faces.append((cx, cy, fw, fh, angle)) |
|
|
| faces.sort(key=lambda x: max(x[2], x[3]), reverse=True) |
| return faces |
|
|
|
|
| def rotate_image(image, angle, cx, cy): |
| h, w = image.shape[0:2] |
| rot_mat = cv2.getRotationMatrix2D((cx, cy), angle, 1.0) |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| result = cv2.warpAffine(image, rot_mat, (w, h), flags=cv2.INTER_LINEAR, borderMode=cv2.BORDER_REFLECT) |
| return result, cx, cy |
|
|
|
|
| def process(args): |
| assert (not args.resize_fit) or args.resize_face_size is None, f"resize_fit and resize_face_size can't be specified both / resize_fitとresize_face_sizeはどちらか片方しか指定できません" |
| assert args.crop_ratio is None or args.resize_face_size is None, f"crop_ratio指定時はresize_face_sizeは指定できません" |
|
|
| |
| logger.info("loading face detector.") |
| detector = create_detector('yolov3') |
|
|
| |
| if args.crop_size is None: |
| crop_width = crop_height = None |
| else: |
| tokens = args.crop_size.split(',') |
| assert len(tokens) == 2, f"crop_size must be 'width,height' / crop_sizeは'幅,高さ'で指定してください" |
| crop_width, crop_height = [int(t) for t in tokens] |
|
|
| if args.crop_ratio is None: |
| crop_h_ratio = crop_v_ratio = None |
| else: |
| tokens = args.crop_ratio.split(',') |
| assert len(tokens) == 2, f"crop_ratio must be 'horizontal,vertical' / crop_ratioは'幅,高さ'の倍率で指定してください" |
| crop_h_ratio, crop_v_ratio = [float(t) for t in tokens] |
|
|
| |
| logger.info("processing.") |
| output_extension = ".png" |
|
|
| os.makedirs(args.dst_dir, exist_ok=True) |
| paths = glob.glob(os.path.join(args.src_dir, "*.png")) + glob.glob(os.path.join(args.src_dir, "*.jpg")) + \ |
| glob.glob(os.path.join(args.src_dir, "*.webp")) |
| for path in tqdm(paths): |
| basename = os.path.splitext(os.path.basename(path))[0] |
|
|
| |
| image = cv2.imdecode(np.fromfile(path, np.uint8), cv2.IMREAD_UNCHANGED) |
| if len(image.shape) == 2: |
| image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR) |
| if image.shape[2] == 4: |
| logger.warning(f"image has alpha. ignore / 画像の透明度が設定されているため無視します: {path}") |
| image = image[:, :, :3].copy() |
|
|
| h, w = image.shape[:2] |
|
|
| faces = detect_faces(detector, image, args.multiple_faces) |
| for i, face in enumerate(faces): |
| cx, cy, fw, fh, angle = face |
| face_size = max(fw, fh) |
| if args.min_size is not None and face_size < args.min_size: |
| continue |
| if args.max_size is not None and face_size >= args.max_size: |
| continue |
| face_suffix = f"_{i+1:02d}" if args.multiple_faces else "" |
|
|
| |
| face_img = image |
| if args.rotate: |
| face_img, cx, cy = rotate_image(face_img, angle, cx, cy) |
|
|
| |
| if crop_width is not None or crop_h_ratio is not None: |
| cur_crop_width, cur_crop_height = crop_width, crop_height |
| if crop_h_ratio is not None: |
| cur_crop_width = int(face_size * crop_h_ratio + .5) |
| cur_crop_height = int(face_size * crop_v_ratio + .5) |
|
|
| |
| scale = 1.0 |
| if args.resize_face_size is not None: |
| |
| scale = args.resize_face_size / face_size |
| if scale < cur_crop_width / w: |
| logger.warning( |
| f"image width too small in face size based resizing / 顔を基準にリサイズすると画像の幅がcrop sizeより小さい(顔が相対的に大きすぎる)ので顔サイズが変わります: {path}") |
| scale = cur_crop_width / w |
| if scale < cur_crop_height / h: |
| logger.warning( |
| f"image height too small in face size based resizing / 顔を基準にリサイズすると画像の高さがcrop sizeより小さい(顔が相対的に大きすぎる)ので顔サイズが変わります: {path}") |
| scale = cur_crop_height / h |
| elif crop_h_ratio is not None: |
| |
| pass |
| else: |
| |
| if w < cur_crop_width: |
| logger.warning(f"image width too small/ 画像の幅がcrop sizeより小さいので画質が劣化します: {path}") |
| scale = cur_crop_width / w |
| if h < cur_crop_height: |
| logger.warning(f"image height too small/ 画像の高さがcrop sizeより小さいので画質が劣化します: {path}") |
| scale = cur_crop_height / h |
| if args.resize_fit: |
| scale = max(cur_crop_width / w, cur_crop_height / h) |
|
|
| if scale != 1.0: |
| w = int(w * scale + .5) |
| h = int(h * scale + .5) |
| if scale < 1.0: |
| face_img = cv2.resize(face_img, (w, h), interpolation=cv2.INTER_AREA) |
| else: |
| face_img = pil_resize(face_img, (w, h)) |
| cx = int(cx * scale + .5) |
| cy = int(cy * scale + .5) |
| fw = int(fw * scale + .5) |
| fh = int(fh * scale + .5) |
|
|
| cur_crop_width = min(cur_crop_width, face_img.shape[1]) |
| cur_crop_height = min(cur_crop_height, face_img.shape[0]) |
|
|
| x = cx - cur_crop_width // 2 |
| cx = cur_crop_width // 2 |
| if x < 0: |
| cx = cx + x |
| x = 0 |
| elif x + cur_crop_width > w: |
| cx = cx + (x + cur_crop_width - w) |
| x = w - cur_crop_width |
| face_img = face_img[:, x:x+cur_crop_width] |
|
|
| y = cy - cur_crop_height // 2 |
| cy = cur_crop_height // 2 |
| if y < 0: |
| cy = cy + y |
| y = 0 |
| elif y + cur_crop_height > h: |
| cy = cy + (y + cur_crop_height - h) |
| y = h - cur_crop_height |
| face_img = face_img[y:y + cur_crop_height] |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| if args.debug: |
| cv2.rectangle(face_img, (cx-fw//2, cy-fh//2), (cx+fw//2, cy+fh//2), (255, 0, 255), fw//20) |
|
|
| _, buf = cv2.imencode(output_extension, face_img) |
| with open(os.path.join(args.dst_dir, f"{basename}{face_suffix}_{cx:04d}_{cy:04d}_{fw:04d}_{fh:04d}{output_extension}"), "wb") as f: |
| buf.tofile(f) |
|
|
|
|
| def setup_parser() -> argparse.ArgumentParser: |
| parser = argparse.ArgumentParser() |
| parser.add_argument("--src_dir", type=str, help="directory to load images / 画像を読み込むディレクトリ") |
| parser.add_argument("--dst_dir", type=str, help="directory to save images / 画像を保存するディレクトリ") |
| parser.add_argument("--rotate", action="store_true", help="rotate images to align faces / 顔が正立するように画像を回転する") |
| parser.add_argument("--resize_fit", action="store_true", |
| help="resize to fit smaller side after cropping / 切り出し後の画像の短辺がcrop_sizeにあうようにリサイズする") |
| parser.add_argument("--resize_face_size", type=int, default=None, |
| help="resize image before cropping by face size / 切り出し前に顔がこのサイズになるようにリサイズする") |
| parser.add_argument("--crop_size", type=str, default=None, |
| help="crop images with 'width,height' pixels, face centered / 顔を中心として'幅,高さ'のサイズで切り出す") |
| parser.add_argument("--crop_ratio", type=str, default=None, |
| help="crop images with 'horizontal,vertical' ratio to face, face centered / 顔を中心として顔サイズの'幅倍率,高さ倍率'のサイズで切り出す") |
| parser.add_argument("--min_size", type=int, default=None, |
| help="minimum face size to output (included) / 処理対象とする顔の最小サイズ(この値以上)") |
| parser.add_argument("--max_size", type=int, default=None, |
| help="maximum face size to output (excluded) / 処理対象とする顔の最大サイズ(この値未満)") |
| parser.add_argument("--multiple_faces", action="store_true", |
| help="output each faces / 複数の顔が見つかった場合、それぞれを切り出す") |
| parser.add_argument("--debug", action="store_true", help="render rect for face / 処理後画像の顔位置に矩形を描画します") |
|
|
| return parser |
|
|
|
|
| if __name__ == '__main__': |
| parser = setup_parser() |
|
|
| args = parser.parse_args() |
|
|
| process(args) |
|
|