tools/accuracy_checker/accuracy_checker/postprocessor/normalize_landmarks_points.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 import numpy as np
  18
  19 from ..config import BoolField
  20 from ..postprocessor.postprocessor import Postprocessor, BasePostprocessorConfig
  21 from ..representation import FacialLandmarksAnnotation, FacialLandmarksPrediction
  22
  23
  24 class NormalizeLandmarksPoints(Postprocessor):
  25     __provider__ = 'normalize_landmarks_points'
  26
  27     annotation_types = (FacialLandmarksAnnotation, )
  28     prediction_types = (FacialLandmarksPrediction, )
  29
  30     def validate_config(self):
  31         class _ConfigValidator(BasePostprocessorConfig):
  32             use_annotation_rect = BoolField(optional=True)
  33
  34         config_validator = _ConfigValidator(
  35             self.__provider__, on_extra_argument=_ConfigValidator.ERROR_ON_EXTRA_ARGUMENT
  36         )
  37         config_validator.validate(self.config)
  38
  39     def configure(self):
  40         self.use_annotation_rect = self.config.get('use_annotation_rect', False)
  41
  42     def process_image(self, annotation, prediction):
  43         for target in annotation:
  44             height, width, _ = self.image_size
  45             x_start, y_start = 0, 0
  46             if self.use_annotation_rect:
  47                 resized_box = annotation[0].metadata.get('rect')
  48                 x_start, y_start, x_max, y_max = resized_box
  49                 width = x_max - x_start
  50                 height = y_max - y_start
  51
  52             target.x_values = (
  53                 (np.array(target.x_values, dtype=float) - x_start) / np.maximum(width, np.finfo(np.float64).eps)
  54             )
  55             target.y_values = (
  56                 (np.array(target.y_values, dtype=float) - y_start) / np.maximum(height, np.finfo(np.float64).eps)
  57             )
  58
  59         return annotation, prediction