def testAffineAlign(self): aligner = AffineAligner(fidu_model_file = '../resources/model_ang_0.txt') img_files = ['./resources/affine_align/Meryl_Streep_0013.jpg', './resources/affine_align/Fayssal_Mekdad_0002.jpg'] for img_file in img_files: img_file = os.path.abspath(img_file) detect_landmarks(fname = img_file) fidu_file = img_file.replace('.jpg','.cfidu') img = cv2.imread(img_file) score, yaw_angle, fidu_points = read_fidu(fidu_file) aligned_img, R = aligner.align(img, fidu_points) aligned_img_cpy = aligned_img.copy() draw_fidu(img, fidu_points) fidu_points_in_aligned = unwarp_fidu(orig_fidu_points = fidu_points, unwarp_mat = R) draw_fidu(aligned_img_cpy, fidu_points_in_aligned, radius = 9, color = (255,0,0), thickness = 3) cv2.imshow('padded_face with landmarks', img) cv2.imshow('aligned_face', cv2.resize(aligned_img, (320,320), interpolation = cv2.INTER_CUBIC)) cv2.imshow('aligned_face with landmarks', cv2.resize(aligned_img_cpy, (320,320), interpolation = cv2.INTER_CUBIC)) cv2.waitKey()
def testAffineAlign(self): aligner = AffineAligner(fidu_model_file='../resources/model_ang_0.txt') img_files = [ './resources/affine_align/Meryl_Streep_0013.jpg', './resources/affine_align/Fayssal_Mekdad_0002.jpg' ] for img_file in img_files: img_file = os.path.abspath(img_file) detect_landmarks(fname=img_file) fidu_file = img_file.replace('.jpg', '.cfidu') img = cv2.imread(img_file) score, yaw_angle, fidu_points = read_fidu(fidu_file) aligned_img, R = aligner.align(img, fidu_points) aligned_img_cpy = aligned_img.copy() draw_fidu(img, fidu_points) fidu_points_in_aligned = unwarp_fidu(orig_fidu_points=fidu_points, unwarp_mat=R) draw_fidu(aligned_img_cpy, fidu_points_in_aligned, radius=9, color=(255, 0, 0), thickness=3) cv2.imshow('padded_face with landmarks', img) cv2.imshow( 'aligned_face', cv2.resize(aligned_img, (320, 320), interpolation=cv2.INTER_CUBIC)) cv2.imshow( 'aligned_face with landmarks', cv2.resize(aligned_img_cpy, (320, 320), interpolation=cv2.INTER_CUBIC)) cv2.waitKey()
class CascadeFaceAligner(object): ''' classdocs ''' def __init__(self, haar_file = '../resources/haarcascade_frontalface_default.xml', lbp_file = '../resources/lbpcascade_frontalface.xml', fidu_model_file = '../resources/model_ang_0.txt', fidu_exec_dir = '../resources/'): ''' Constructor ''' self.face_finder = CascadeFaceFinder(haar_file = haar_file, lbp_file = lbp_file) self.aligner = AffineAligner(fidu_model_file = fidu_model_file) self.fidu_exec_dir = fidu_exec_dir self.valid_angles = [-45,-30,-15,0,15,30,45] def detect_faces(self, input_folder, output_folder, mark_dones = True): ''' mark_dones - if True, will create a hidden file, marking this file as done with a hidden file, starting with '.done.' ''' input_files1 = glob.glob(os.path.join(input_folder, '*.jpg')) input_files2 = glob.glob(os.path.join(input_folder, '*.png')) input_files = input_files1 + input_files2 N = len(input_files) for n_file, input_file in enumerate(input_files): a,b = os.path.split(input_file) done_file = os.path.join(a, '.done.' + b.rsplit('.',1)[0]) if os.path.exists(done_file): continue print ("... processing", input_file) target_faces_file = os.path.join( output_folder, os.path.split(input_file)[1].rsplit('.',1)[0] + '.faces.txt') faces_file = self.face_finder.create_faces_file( input_file, is_overwrite = False, target_file = target_faces_file ) sub_images_files = self.face_finder.create_sub_images_from_file( original_image_file = input_file, faces_file = faces_file, target_folder = output_folder, img_type = 'jpg') #touch open(done_file,'w').close() print ("Detected on %d / %d files" %(n_file, N)) def align_faces(self, input_images, output_path, fidu_max_size = None, fidu_min_size = None, is_align = True, is_draw_fidu = False, delete_no_fidu = False): ''' input_images - can be either a folder (all *.jpgs in it) or a list of filenames , fidu_max_size = None, fidu_min_size = None): ''' if type(input_images) == type(''): input_images = glob.glob(os.path.join(input_images, '*.jpg')) for input_image in input_images: detect_landmarks(fname = os.path.abspath(input_image), max_size = fidu_max_size, min_size = fidu_min_size, fidu_exec_dir = self.fidu_exec_dir) fidu_file = input_image.rsplit('.',1)[0] + '.cfidu' fidu_score, yaw_angle, fidu_points = read_fidu(fidu_file) if not (fidu_score is not None and yaw_angle in self.valid_angles): # skip face if delete_no_fidu: os.remove(fidu_file) os.remove(input_image) continue if is_align: # save the aligned image sub_img = cv2.imread(input_image) _, base_fname = os.path.split(input_image) aligned_img, R = self.aligner.align(sub_img, fidu_points) aligned_img_file = os.path.join(output_path, base_fname.rsplit('.',1)[0] + '.aligned.png') cv2.imwrite(aligned_img_file, aligned_img) # save a copy of the aligned image, with the landmarks drawn if is_draw_fidu: aligned_img_file = os.path.join(output_path, base_fname.rsplit('.',1)[0] + '.aligned.withpoints.png') fidu_points_in_aligned = unwarp_fidu(orig_fidu_points = fidu_points, unwarp_mat = R) draw_fidu(aligned_img, fidu_points_in_aligned, radius = 9, color = (255,0,0), thickness = 3) cv2.imwrite(aligned_img_file, aligned_img)
class CascadeFaceAligner(object): ''' classdocs ''' def __init__(self, haar_file = '../resources/haarcascade_frontalface_default.xml', lbp_file = '../resources/lbpcascade_frontalface.xml', fidu_model_file = '../resources/model_ang_0.txt', fidu_exec_dir = '../resources/'): ''' Constructor ''' self.face_finder = CascadeFaceFinder(haar_file = haar_file, lbp_file = lbp_file) self.aligner = AffineAligner(fidu_model_file = fidu_model_file) self.fidu_exec_dir = fidu_exec_dir self.valid_angles = [-45,-30,-15,0,15,30,45] def detect_faces(self, input_folder, output_folder, mark_dones = True): ''' mark_dones - if True, will create a hidden file, marking this file as done with a hidden file, starting with '.done.' ''' input_files1 = glob.glob(os.path.join(input_folder, '*.jpg')) input_files2 = glob.glob(os.path.join(input_folder, '*.png')) input_files = input_files1 + input_files2 N = len(input_files) for n_file, input_file in enumerate(input_files): a,b = os.path.split(input_file) done_file = os.path.join(a, '.done.' + b.rsplit('.',1)[0]) if os.path.exists(done_file): continue print "... processing", input_file target_faces_file = os.path.join( output_folder, os.path.split(input_file)[1].rsplit('.',1)[0] + '.faces.txt') faces_file = self.face_finder.create_faces_file( input_file, is_overwrite = False, target_file = target_faces_file ) sub_images_files = self.face_finder.create_sub_images_from_file( original_image_file = input_file, faces_file = faces_file, target_folder = output_folder, img_type = 'jpg') #touch open(done_file,'w').close() print "Detected on %d / %d files" %(n_file, N) def align_faces(self, input_images, output_path, fidu_max_size = None, fidu_min_size = None, is_align = True, is_draw_fidu = False, delete_no_fidu = False): ''' input_images - can be either a folder (all *.jpgs in it) or a list of filenames , fidu_max_size = None, fidu_min_size = None): ''' if type(input_images) == type(''): input_images = glob.glob(os.path.join(input_images, '*.jpg')) for input_image in input_images: detect_landmarks(fname = os.path.abspath(input_image), max_size = fidu_max_size, min_size = fidu_min_size, fidu_exec_dir = self.fidu_exec_dir) fidu_file = input_image.rsplit('.',1)[0] + '.cfidu' fidu_score, yaw_angle, fidu_points = read_fidu(fidu_file) if not (fidu_score is not None and yaw_angle in self.valid_angles): # skip face if delete_no_fidu: os.remove(fidu_file) os.remove(input_image) continue if is_align: # save the aligned image sub_img = cv2.imread(input_image) _, base_fname = os.path.split(input_image) aligned_img, R = self.aligner.align(sub_img, fidu_points) aligned_img_file = os.path.join(output_path, base_fname.rsplit('.',1)[0] + '.aligned.png') cv2.imwrite(aligned_img_file, aligned_img) # save a copy of the aligned image, with the landmarks drawn if is_draw_fidu: aligned_img_file = os.path.join(output_path, base_fname.rsplit('.',1)[0] + '.aligned.withpoints.png') fidu_points_in_aligned = unwarp_fidu(orig_fidu_points = fidu_points, unwarp_mat = R) draw_fidu(aligned_img, fidu_points_in_aligned, radius = 9, color = (255,0,0), thickness = 3) cv2.imwrite(aligned_img_file, aligned_img)