tools/accuracy_checker/accuracy_checker/annotation_converters/lfw.py

   1 """
   2 Copyright (c) 2019 Intel Corporation
   3
   4 Licensed under the Apache License, Version 2.0 (the "License");
   5 you may not use this file except in compliance with the License.
   6 You may obtain a copy of the License at
   7
   8       http://www.apache.org/licenses/LICENSE-2.0
   9
  10 Unless required by applicable law or agreed to in writing, software
  11 distributed under the License is distributed on an "AS IS" BASIS,
  12 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 See the License for the specific language governing permissions and
  14 limitations under the License.
  15 """
  16
  17 from collections import defaultdict
  18 from pathlib import Path
  19
  20 from ..config import PathField
  21 from ..representation import ReIdentificationClassificationAnnotation
  22 from ..utils import read_txt
  23
  24 from .format_converter import BaseFormatConverter, BaseFormatConverterConfig
  25
  26
  27 class FaceReidPairwiseConverterConfig(BaseFormatConverterConfig):
  28     pairs_file = PathField()
  29     train_file = PathField(optional=True)
  30     landmarks_file = PathField(optional=True)
  31
  32
  33 class FaceReidPairwiseConverter(BaseFormatConverter):
  34     __provider__ = 'face_reid_pairwise'
  35
  36     _config_validator_type = FaceReidPairwiseConverterConfig
  37
  38     def configure(self):
  39         self.pairs_file = self.config['pairs_file']
  40         self.train_file = self.config.get('train_file')
  41         self.landmarks_file = self.config.get('landmarks_file')
  42
  43     def convert(self):
  44         landmarks_map = {}
  45         if self.landmarks_file:
  46             for landmark_line in read_txt(self.landmarks_file):
  47                 landmark_line = landmark_line.split('\t')
  48                 landmarks_map[landmark_line[0]] = [int(point) for point in landmark_line[1:]]
  49
  50         test_annotations = self.prepare_annotation(self.pairs_file, True, landmarks_map)
  51         if self.train_file:
  52             train_annotations = self.prepare_annotation(self.train_file, True, landmarks_map)
  53             test_annotations += train_annotations
  54
  55         return test_annotations, None
  56
  57     @staticmethod
  58     def get_image_name(person, image_id):
  59         image_path_pattern = '{}/{}_{}{}.jpg'
  60         return image_path_pattern.format(person, person, '0' * (4 - len(image_id)), image_id)
  61
  62     def convert_positive(self, pairs, all_images):
  63         positives = defaultdict(set)
  64         for data in pairs:
  65             image1 = self.get_image_name(data[0], data[1])
  66             image2 = self.get_image_name(data[0], data[2])
  67             positives[image1].add(image2)
  68             all_images.add(image1)
  69             all_images.add(image2)
  70
  71         return positives, all_images
  72
  73     def convert_negative(self, pairs, all_images):
  74         negatives = defaultdict(set)
  75         for data in pairs:
  76             image1 = self.get_image_name(data[0], data[1])
  77             image2 = self.get_image_name(data[2], data[3])
  78             negatives[image1].add(image2)
  79             all_images.add(image1)
  80             all_images.add(image2)
  81
  82         return negatives, all_images
  83
  84     def prepare_annotation(self, ann_file: Path, train=False, landmarks_map=None):
  85         positive_pairs, negative_pairs = [], []
  86         ann_lines = read_txt(ann_file)
  87         for line in ann_lines[1:]:  # skip header
  88             pair = line.strip().split()
  89             if len(pair) == 3:
  90                 positive_pairs.append(pair)
  91             elif len(pair) == 4:
  92                 negative_pairs.append(pair)
  93
  94         all_images = set()
  95         positive_data, all_images = self.convert_positive(positive_pairs, all_images)
  96         negative_data, all_images = self.convert_negative(negative_pairs, all_images)
  97
  98         annotations = []
  99         for image in all_images:
 100             annotation = ReIdentificationClassificationAnnotation(image, positive_data[image], negative_data[image])
 101
 102             if landmarks_map:
 103                 image_landmarks = landmarks_map.get(image)
 104                 annotation.metadata['keypoints'] = image_landmarks
 105
 106             if train:
 107                 annotation.metadata['train'] = True
 108
 109             annotations.append(annotation)
 110
 111         return annotations