Python 動作確認

Question

python3を使いネットで拾ったコードをほんの少し自分なりに変形して実行しているのですが、for文が回らない箇所がいくつかあります。

回っていないfor文は

for item in filtered:
for item in finals:

です。原因がわかりません。

import numpy as np
import sys
import os
import math
import cv2
import glob

IMAGE_SIZE = 112
INPUT_SIZE = 96

xml_dir = '/usr/local/Cellar/opencv3/3.2.0/share/OpenCV/haarcascades'
face_cascade = cv2.CascadeClassifier(os.path.join(xml_dir, 'haarcascade_frontalface_alt2.xml'))
eye_cascade = cv2.CascadeClassifier(os.path.join(xml_dir, 'haarcascade_eye.xml'))
mouth_cascade = cv2.CascadeClassifier(os.path.join(xml_dir, 'haarcascade_mcs_mouth.xml'))
nose_cascade = cv2.CascadeClassifier(os.path.join(xml_dir, 'haarcascade_mcs_nose.xml'))


def detect_face_rotate(img_file, out_dir = 'out'):
    base_dir = "."
    #out_dir = "out"
    filename = os.path.basename(os.path.normpath(img_file))
    (fn, ext) = os.path.splitext(filename)
    img_dir = '%s/%s/%s' % (".", out_dir, fn)
    
    
    input_img = cv2.imread(img_file)
    rows, cols, colors = input_img.shape

    print ('size',rows,cols)
    limit = 800
    if rows > limit or cols > limit:
        if rows > cols:
            input_img = cv2.resize(input_img, (cols * limit / rows, limit))
        else:
            input_img = cv2.resize(input_img, (int(limit), int(rows * limit / cols)))
        rows, cols, colors = input_img.shape
        print ('resized', rows, cols)

    gray = cv2.cvtColor(input_img, cv2.COLOR_BGR2GRAY)
    hypot = int(math.hypot(rows, cols))
    frame = np.zeros((hypot, hypot), np.uint8)
    frame[int((hypot - rows) * 0.5):int((hypot + rows) * 0.5), int((hypot - cols) * 0.5):int((hypot + cols) * 0.5)] = gray
    results = []
    face_id_seed = 0

    #5度ずつ元画像を回転し、顔の候補を全部取得
    #for deg in range(-50, 51, 5):
    for deg in range(-5, 5, 5):
        print('deg:%s' % deg)
        M = cv2.getRotationMatrix2D((hypot * 0.5, hypot * 0.5), -deg, 1.0)
        rotated = cv2.warpAffine(frame, M, (hypot, hypot))
        #"""
        out_file = '%s/deg_%s.jpg' % (img_dir, deg)

        faces = face_cascade.detectMultiScale(rotated, 1.02, 5)
        for (x, y, w, h) in faces:
            
            face_cand = rotated[y:y+h, x:x+w]

            center = (int(x + w * 0.5), int(y + h * 0.5))
            origin = (int(hypot * 0.5), int(hypot * 0.5))
            r_deg = -deg
            center_org = rotate_coord(center, origin, r_deg)
            print ('face', (x,y,w,h), center_org)

            resized = face_cand
            if w < IMAGE_SIZE:
                print ('resizing..')
                resized = cv2.resize(face_cand, (IMAGE_SIZE, IMAGE_SIZE))

            result = {
                    'face_id': 'f%s' % face_id_seed,
                    'img_resized': resized, #顔候補bitmap(小さい場合zoom)
                    'img': face_cand, #顔候補bitmap(元サイズ)
                    'deg': deg, #回転
                    'frame': (x, y, w, h), #回転状態における中心座標+size
                    'center_org': center_org, #角度0時における中心座標
                    }
            results.append(result)
            face_id_seed += 1

    eyes_id_seed = 0
    eyes_faces = []

    for result in results:
        print ('#eyes:',result['face_id'])

        img = np.copy(result["img_resized"])
        fw,fh = img.shape
        eyes = eye_cascade.detectMultiScale(img, 1.02)
        left_eye = right_eye = None #左上/右上にそれぞれ目が１つずつ検出できればOK
        for (x,y,w,h) in eyes:
           
            print ('## eye:',x,y,w,h,fw/8,fw/4,fh/8,fh/4)
            cv2.rectangle(img,(x,y),(x+w,y+h),(64,64,0),1)

            if not (fw/6 < w and w < fw/2):
                print ('eye width invalid')
                continue
            if not (fh/6 < h and h < fh/2):
                print ('eye height invalid')
                continue
            if not fh * 0.5 - (y + h * 0.5) > 0: #上半分に存在する
                print ('eye position invalid')
                continue
            if fw * 0.5 - (x + w * 0.5) > 0:
                if left_eye:
                    print ('too many left eye')
                    continue
                else:
                    left_eye = (x,y,w,h)
            else:
                if right_eye:
                    print ('too many right eye')
                    continue
                else:
                    right_eye = (x,y,w,h)
        if left_eye and right_eye:
            print ('>>> valid eyes detect')
            result['left_eye'] = left_eye
            result['right_eye'] = right_eye
            eyes_faces.append(result)

    #重複検出を除去
    candidates = []
    for i, result in enumerate(eyes_faces):
        print ('result:',result['face_id'])
        result['duplicated'] = False
        for cand in candidates:
            c_x, c_y = cand['center_org']
            _,_,cw,ch = cand['frame']
            r_x, r_y = result['center_org']
            _,_,rw,rh = result['frame']
            if abs(c_x - r_x) < ((cw+rw)*0.5*0.3) and abs(c_y - r_y) < ((ch+rh)*0.5*0.3): #近い場所にある顔候補
                c_diff = eyes_vertical_diff(cand)
                r_diff = eyes_vertical_diff(result)
                print ('c_diff:',cand['face_id'],c_diff)
                print ('r_diff:',result['face_id'],r_diff)
                if c_diff < r_diff: #より左右の目の水平位置が近いほうが採用
                    result['duplicated'] = True
                else:
                    cand['duplicated'] = True
        candidates.append(result)
    filtered = filter(lambda n: n['duplicated'] == False, candidates)

    #"""
    for cand in filtered:
       
        out_file = '%s/cand_%s_face_%s_%s.jpg' % (img_dir, cand['face_id'], cand['deg'], cand['center_org'])
        #cv2.imwrite(out_file, cand['img'])
        #res.append(out_file)
    #"""
    finals = []
    #候補に対してさらに口検出チェック
    for item in filtered:
        img = np.copy(item["img_resized"])
        fw,fh = img.shape
        mouthes = mouth_cascade.detectMultiScale(img, 1.02) #faceの中心下部付近にあればOK
        mouth_found = False
        for (mx,my,mw,mh) in mouthes:
            
            print ('mouth',(mx,my,mw,mh))
            cv2.rectangle(img,(mx,my),(mx+mw,my+mh),(128,128,0),2)
            h_diff = fh/2 - (my+mh/2)
            print (fh, h_diff)
            if h_diff < 0:
                mouth_found = True
                break
        #"""
        out_file = '%s/mouth_%s_face_%s_%s.jpg' % (img_dir, item['face_id'], item['deg'], item['center_org'])
        #cv2.imwrite(out_file, img)
        #"""
        print (item['face_id'], 'mouth found?', mouth_found)
        if mouth_found:
            finals.append(item)
            out_file1 = '%s/final_%s_%s_%s_%s.jpg' % (out_dir, fn, item['face_id'], item['deg'], item['center_org'])
            out_file2 = '%s/final_%s_%s_%s_%s.jpg' % (img_dir, fn, item['face_id'], item['deg'], item['center_org'])
            #cv2.imwrite(out_file1, item['img'])
            #cv2.imwrite(out_file2, item['img'])

    #最後にカラー画像として切り出し
    res = []
    for item in finals:
        
        out_file = crop_color_face(item, input_img, base_dir, out_dir, fn)
        res.append(out_file)
    return res

def crop_color_face(item, img, base_dir, out_dir, fn):
    rows, cols, colors = img.shape
    hypot = int(math.hypot(rows, cols))
    frame = np.zeros((hypot, hypot, 3), np.uint8)
    frame[int((hypot - rows) * 0.5):int((hypot + rows) * 0.5), int((hypot - cols) * 0.5):int((hypot + cols) * 0.5)] = img

    deg = item['deg']
    M = cv2.getRotationMatrix2D((hypot * 0.5, hypot * 0.5), -deg, 1.0)
    rotated = cv2.warpAffine(frame, M, (hypot, hypot))

    #"""
    #out_file = '%s/7_color_deg_%s.jpg' % (img_dir, deg)
    #cv2.imwrite(out_file, rotated)
    #"""

    x,y,w,h = item['frame']
    face = rotated[y:y+h, x:x+w]
    face = cv2.resize(face, (IMAGE_SIZE, IMAGE_SIZE))
    #"""
    web_path = '%s/%s_%s.jpg' % (out_dir, fn, item['face_id'])
    out_file = '%s/%s' % (base_dir, web_path)
    print ('web_path', web_path)
    print ('out_file', out_file)
    cv2.imwrite(out_file, face)
    #"""

    return web_path

def eyes_vertical_diff(face):
    _,ly,_,lh = face["left_eye"]
    _,ry,_,rh = face["right_eye"]
    return abs((ly + lh * 0.5) - (ry + rh * 0.5))

def rotate_coord(pos, origin, deg):
    """
    posをdeg度回転させた座標を返す
    pos: 対象となる座標tuple(x,y)
    origin: 原点座標tuple(x,y)
    deg: 回転角度
    @return: 回転後の座標tuple(x,y)
    @see: http://www.geisya.or.jp/~mwm48961/kou2/linear_image3.html
    """
    x, y = pos
    ox, oy = origin
    r = np.radians(deg)
    xd = ((x - ox) * np.cos(r) - (y - oy) * np.sin(r)) + ox
    xy = ((x - ox) * np.sin(r) + (y - oy) * np.cos(r)) + oy
    return (int(xd), int(xy))

def show(img):
    cv2.imshow('img',img)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
if __name__ == "__main__":
    param = sys.argv
    
    img_dir = param[1]
    dir_check = os.path.exists(img_dir+"_out_img")#指定の画像格納ディレクトリの有無
    #無ければ作成
    if dir_check == False:
        os.mkdir(img_dir+"_out_img")
        
    #os.chdir(dir)
    #ファイル読み込み
    img_file = glob.glob("./"+img_dir+"/*")
    outdir = img_dir+"_out_img"
    
    for img in img_file:
        detect_face_rotate(img,outdir)

Comments

[yukkeorg]

質問を最後まで記載してください、また、実行した時にエラーが出る場合は、そのまま記載をお願いします、

[daigakuse-]

誤字訂正しました。エラーはないです。

Answer 1

Python3に組み込みのfilter関数はイテレータをを返します。イテレータは一度ループで処理(for cond in filterd:)してしまうと巻き戻りが起きないため、その後のループ(for item in filtered:)で利用しようとしてもイテレータがさすデータがないことから、ループ処理が行われません。

今回の場合は、以下のようにfilter関数で帰ってきたイテレータで順次取得できる値をリストすると良いかと思います。

    filtered = list(filter(lambda n: n['duplicated'] == False, candidates))

finalsのほうは上のfilteredのループが動いていないため、空リストとなっています。